The economy today, as SarthakGPT reads it

In September 2026 the activity factor is 0.45, meaning the combined monthly data show growth a bit under half a standard deviation faster than the average pace since 1990. A reading of zero would mean growth exactly at its usual pace, and negative readings mean slower than usual.

Simple exampleYou can tell how busy a shopping mall is from the full parking lot, the food court line and the crowded escalators, even without counting people. This gauge does the same with many monthly economic reports. Zero means the usual pace of growth, above zero faster than usual, below zero slower.

Model noteIn Sep 2026 the activity factor is estimated 0.44 standard deviations above its average (trend) growth pace.

Typical rangeBetween -0.475 and 0.393 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -0.914 to 0.770). The latest reading sits at the 79th percentile.

Averaged over the last three months, the factor is 0.31 above zero, so the recent stretch is above its usual pace but less strong than September alone. The model's own uncertainty around this average is 0.29, which is close to the size of the reading itself.

Simple exampleOne hot day does not make a heatwave, so this averages the activity gauge over three months for a steadier picture. The uncertainty figure is like the margin of error on a poll: when it is about as big as the reading, the reading is a rough sign, not a precise measure.

Model noteOver the last three months the factor has averaged 0.30 s.d. above trend; the model's state uncertainty around that average is 0.29 s.d. (parameters held fixed).

Typical rangeBetween -0.267 and 0.207 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -0.592 to 0.455). The latest reading sits at the 83th percentile.

The model puts a 14.5% chance on the three-month average actually being below its usual pace. This counts only uncertainty about the hidden factor with the model taken as given, so the true uncertainty is larger than this number suggests.

Simple exampleThis is the model's estimated chance that the economy has actually been growing slower than usual over the last three months, despite what the gauge shows. It only counts the model's own measuring fuzz, so like a bathroom scale that may itself be slightly off, the real uncertainty is larger.

Model noteGiven the model's state uncertainty alone there is a 14.6% probability that the factor averaged below trend over the last three months; this excludes parameter and model uncertainty and is therefore overly precise.

For the July to September 2026 quarter, which has ended but whose official GDP figure is not yet out, the model estimates growth of 3.3% at an annual rate, with a wide likely range of 1.1% to 5.6% (a 68% band, roughly a two-in-three chance range). Over the last 8 released quarters the model missed by 1.24 points on average, more than the 1.09 points from simply using the long-run average growth rate, so this estimate deserves caution.

Simple examplePicture a game that has ended, but the official score has not been posted yet. Using the highlights already seen, the model estimates the final score: here, how fast the economy grew last quarter. The wide range shows the guess can be off by quite a bit.

Model noteThe model puts 2026Q3 real GDP growth at 3.5% annualised (68% band 1.9 to 5.1, data through Sep 2026), but over the last 8 released quarters the held-out model nowcast was less accurate than simply using the long-run average growth rate (MAE 1.25 vs 1.09 pp).

Typical rangeBetween 1.25 and 3.51 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -0.588 to 4.58). The latest reading sits at the 74th percentile.

For the current quarter (October to December 2026), the model estimates 3.1% annual-rate growth, with a 68% band of 0.5% to 5.8%. This uses data only through September, so it is mostly a projection, and the same track-record caveat applies (average miss 1.24 points versus 1.09 for the long-run average).

Simple exampleThis estimates growth for the quarter happening right now, like guessing a game's final score at halftime. Most of the game has not been played yet, so it leans heavily on how things were going, and the range around it is wide.

Model noteThe model puts 2026Q4 real GDP growth at 3.1% annualised (68% band 1.0 to 5.3, data through Sep 2026), but over the last 8 released quarters the held-out model nowcast was less accurate than simply using the long-run average growth rate (MAE 1.25 vs 1.09 pp).

Typical rangeBetween 1.25 and 3.51 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -0.588 to 4.58). The latest reading sits at the 63th percentile.

The newest industrial production report pushed the September factor up by 0.30, the largest single push from any series. All of the newest reports together moved the factor up by 0.40.

Simple exampleWhen a soup tastes different, you can ask which ingredient changed it most. This names the newest report that moved the activity gauge the most, and by how much, along with the total push from all new reports combined.

Model noteThe latest industrial production print moved the Sep 2026 factor by +0.30 s.d. (total update from all latest prints +0.42).

Core PCE prices, which leave out food and energy, are 3.01% higher than a year ago. That is 1.01 percentage points above the Fed's 2% target.

Simple exampleIf a basket of everyday things (leaving out food and fuel, which jump around a lot) cost $100 a year ago and $103 today, prices rose 3%. This number is that yearly rise. The Fed aims for about 2%, so a higher number means prices are climbing faster than it wants.

Model noteCore PCE prices are 3.01% above a year ago, +1.01pp versus the Fed's 2% target.

Typical rangeBetween 1.54 and 2.29 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 1.28 to 3.22). The latest reading sits at the 87th percentile.

Over the last three months core PCE prices rose at a 2.05% annual pace, 0.96 points below the 12-month rate. A recent pace below the yearly pace means inflation momentum is slowing.

Simple exampleA runner's average speed over a whole race can differ from their speed on the last few laps. This is the price rise over just the last three months, stretched to a yearly pace. If it sits below the yearly figure, price growth has been slowing lately; above it, speeding up.

Model noteOver the last three months core PCE ran at a 2.05% annualized pace, below the 12-month rate by 0.96pp (decelerating momentum).

Typical rangeBetween 1.45 and 2.57 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 1.13 to 3.56). The latest reading sits at the 56th percentile.

Blending five inflation measures into one common trend puts underlying inflation at 2.22%, modestly above the 2% target. Because the headline core PCE figure is 0.79 points above this broad trend, the extra inflation appears concentrated in a narrower set of prices rather than spread widely.

Simple exampleAsk five thermometers in the same room and each reads a bit differently. This blends five ways of measuring price rises and keeps the part they agree on, giving a steadier sense of the true trend. If the main measure is well above it, the extra may be coming from just a few items.

Model noteThe common component of five core/trimmed/median/sticky measures puts underlying inflation at 2.22% in core-PCE units (modestly above the 2% target), so reported core PCE is 0.79pp above the broad trend: the overshoot is narrow, not shared by the trimmed/median/sticky measures.

Typical rangeBetween 1.48 and 2.23 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 1.14 to 3.12). The latest reading sits at the 74th percentile.

Bond market prices imply average inflation of 2.33% for the period 5 to 10 years from now, close to the roughly 2.25% that fits a 2% Fed target. By the model's yardstick this counts as anchored, though its z-score of +0.98 sits just under the threshold of 1, so it is near the edge of that range.

Simple exampleBond prices let you read what investors expect prices to do far ahead, here the average yearly rise from 5 to 10 years from now. It is like asking people how much they think a movie ticket will cost in a decade. Expectations close to the Fed's goal are called anchored.

Model noteMarket-implied CPI inflation 5-10 years ahead is 2.33%, +0.08pp from the ~2.25% consistent with a 2% PCE target; on the 2026-10 month-to-date average (11 trading days) its z vs the prior 10 years is +0.98 and its distance from 2.25% is 0.09pp, so long-run expectations look anchored, though close to the |z| = 1 threshold (anchored = |z| < 1 and distance < 0.25pp).

Typical rangeBetween 2.09 and 2.46 in half of all readings since Jan 2, 2003 (middle 50%; the outer 10th to 90th percentile band is 1.81 to 2.61). The latest reading sits at the 56th percentile.

Using a neutral rate of 1.09%, core inflation of 3.01% and an output gap of +0.58%, the balanced-approach rule suggests a policy rate of 5.19%. This is a mechanical benchmark, not a recommendation or a forecast of Fed decisions.

Simple exampleThink of a recipe for setting the Fed's interest rate: start from a neutral setting, add current inflation, then add more if prices rise faster than 2% or the economy runs hot. This number is what that recipe gives. It is a simple yardstick, not what the Fed must or will do.

Model noteWith HLW r* = 1.09%, core PCE at 3.01% and an Okun output gap of +0.58%, the balanced-approach rule prescribes a 5.19% policy rate.

Typical rangeBetween -0.364 and 5.17 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -5.54 to 7.51). The latest reading sits at the 76th percentile.

The actual fed funds rate (3.63% average for the month) is 1.56 points below the balanced rule's suggestion. By that rule, policy is looser than the formula would set it.

Simple exampleThis compares the Fed's actual interest rate with what the simple recipe suggests. Like comparing a thermostat's setting to what a rule of thumb says it should be: below the recipe means policy is looser than the rule would set it, above means tighter.

Model noteThe effective fed funds rate (3.63% monthly average) is 1.56pp below the balanced-approach prescription, i.e. policy is looser than that rule.

Typical rangeBetween -1.89 and 0.647 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -2.80 to 5.63). The latest reading sits at the 36th percentile.

After subtracting expected inflation, the fed funds rate is 0.90%, just 0.10 points below the estimated neutral rate of 1.01%. On this gauge policy is roughly neutral, neither clearly pressing on the economy nor clearly boosting it.

Simple exampleThe neutral rate is like coasting on a flat road: neither pressing the gas nor the brake on the economy. This takes the Fed's rate, removes expected inflation, and compares it to that neutral level. Above neutral leans toward braking, below leans toward pressing the gas, and near zero is close to coasting.

Model noteThe real fed funds rate (vs Cleveland Fed 1-yr expected inflation) is 0.90%, 0.10pp below HLW r* (1.01%), so policy is roughly neutral on this gauge (fed funds is a month-to-date average).

Typical rangeBetween -2.69 and -0.641 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -3.57 to 0.572). The latest reading sits at the 84th percentile.

The 10-year US Treasury yields 5.10%. This is the interest rate the US government pays to borrow for 10 years, and many long-term loan rates move with it.

Simple exampleWhen you lend a friend $100 for a year and they promise $105 back, the extra $5 is a 5% interest rate. This number is that rate for lending to the US government for 10 years. When it is higher, mortgages and other long loans usually cost more too.

Model noteThe 10-year Treasury (par CMT) yields 5.10%.

Typical rangeBetween 2.26 and 4.34 in half of all readings since Jan 3, 2000 (middle 50%; the outer 10th to 90th percentile band is 1.68 to 4.92). The latest reading sits at the 93th percentile.

The 10-year rate is 82 basis points (0.82 percentage points) above the 3-month rate, so the yield curve slopes upward. An upward slope is the usual shape; an inverted curve, with short rates above long rates, has often come before recessions.

Simple exampleUsually, tying your money up for longer earns you more, just as a 10-year savings deal normally pays more than a 3-month one. This gap compares the two. A positive gap is the normal shape; when short loans pay more than long ones, that unusual flip has often shown up before past recessions.

Model noteThe curve is upward sloping by 82 bp between 3 months and 10 years.

Typical rangeBetween 0.300 and 2.44 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is -0.363 to 3.18). The latest reading sits at the 36th percentile.

The fitted curve's overall level is 5.45%, its slope is 1.25 points and its mid-range bend is -0.46 points. The smooth curve misses actual yields by 8.6 basis points on average, so it describes the curve closely.

Simple examplePicture interest rates for every loan length plotted as dots, then a smooth line drawn through them. Three dials shape that line: how high it sits, how steeply it tilts, and whether it bulges in the middle. The miss size tells you how closely the smooth line hugs the real dots.

Model noteDiebold-Li factors: level 5.45%, slope (-beta2) 1.25 pp, curvature (beta3) -0.46 pp; fit RMSE 8.6 bp.

Typical rangeBetween 3.01 and 5.00 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is 2.39 to 5.70). The latest reading sits at the 85th percentile.

Japan's 10-year government bond yields 2.95%. Its curve also slopes upward, with the 10-year rate 1.30 points above the 1-year rate.

Simple exampleThis is the same idea as the US 10-year rate, but for lending to Japan's government for 10 years. Comparing countries is like comparing rent in two cities: it shows where borrowing for a long time costs more or less.

Model noteThe 10-year JGB yields 2.95%; the 10y-1y slope is 1.30 pp and NS -beta2 is 1.52 pp.

Typical rangeBetween 0.257 and 1.44 in half of all readings since Jan 4, 2000 (middle 50%; the outer 10th to 90th percentile band is 0.026 to 1.71). The latest reading sits at the 100th percentile.

By the ACM estimate, lenders get about 0.75% a year of extra return for lending to the government for 10 years instead of rolling over short-term loans. A second model, Kim-Wright, puts this term premium at 1.07%, a reminder that it is estimated, not observed.

Simple exampleLending for 10 years in one go is riskier than lending for one year and renewing each year, since a lot can change. The term premium is the extra pay lenders ask for that commitment, a bit like a longer phone contract sometimes costing more per month. It can even be negative. It is estimated, not seen directly, so different models disagree.

Model noteInvestors demand 0.75% a year of compensation for holding 10-year duration risk (ACM); Kim-Wright shows 1.07% as of 2026-10-09.

Typical rangeBetween -0.201 and 1.59 in half of all readings since Jan 3, 2000 (middle 50%; the outer 10th to 90th percentile band is -0.652 to 2.27). The latest reading sits at the 59th percentile.

Over the last three months the ACM term premium rose about 7 basis points. Of the +47 basis point rise in ACM's fitted 10-year yield, about +40 came from higher expected short-term rates and only +7 from the term premium, so most of the move reflects expectations about policy rates.

Simple exampleA long-term rate has two ingredients: what people expect short-term rates to be along the way, and the extra pay for committing long. This splits a recent move in the 10-year rate into those two pieces, like working out whether a bigger grocery bill came from pricier milk or a bigger basket.

Model noteSince 2026-07-15 ACM's fitted 10y zero-coupon yield moved +47 bp (10y par CMT: +55 bp): +40 bp from expected short rates and +7 bp from the term premium.

Typical rangeBetween -25.29 and 21.95 in half of all readings since Jan 3, 2000 (middle 50%; the outer 10th to 90th percentile band is -45.44 to 50.57). The latest reading sits at the 62th percentile.

Using only the slope of the yield curve, the model puts the chance of being in a recession in September 2027 at about 9%, with a 90% range of 4% to 13%. The New York Fed's similar model gives 13%.

Simple exampleLike a weather forecast saying 30% chance of rain, this is the model's estimated chance that the economy will be in a recession about a year from now, judged only from the gap between long and short interest rates. A higher number means the model sees more warning signs. It is an estimate, not a certainty.

Model noteThe term spread alone implies a 9% probability of recession in Sep 2027 (90% CI 4-13%; NY Fed model 13%).

Typical rangeBetween 1.54 and 15.79 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 0.480 to 26.61). The latest reading sits at the 61th percentile.

Adding credit market stress and the inflation-adjusted Fed rate, the model puts the chance of a recession starting within the next 12 months at about 4%, with a 90% range of 0% to 7%. The Fed's own published estimate is 9%, so different models disagree by several points even though all are low.

Simple exampleThis forecast uses more clues than the slope alone: how nervous lenders are about lending to companies, and how high the Fed's rate is after inflation. Think of a doctor checking temperature, pulse and blood pressure instead of just one. Different forecasters often give somewhat different chances, and all are estimates.

Model noteSpread, excess bond premium and real policy rate imply a 4% probability of recession within 12 months (90% CI 0-7%; Fed's published estimate 9%).

Typical rangeBetween 0.785 and 26.59 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 0.133 to 59.26). The latest reading sits at the 43th percentile.

In 1 out of 20 situations like today, average real GDP growth over the next year would come in at or below about 1.17% a year, according to the model. This is the Growth-at-Risk figure, and since it is still positive, the model's bad-luck case is weak growth rather than shrinking output.

Simple exampleWeather forecasts sometimes give a cold-snap scenario as well as a typical day. This is the economy's version: in roughly 1 out of 20 situations like today, average growth over the next year would be this low or lower, by the model's estimate. A positive number means even that bad case is slow growth, not shrinking.

Model noteIn 1 of 20 cases like today, average real GDP growth over the next year would be at or below this (model-implied, NFCI + GDP growth).

Typical rangeBetween 0.471 and 1.33 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -0.129 to 1.79). The latest reading sits at the 65th percentile.

The middle (median) estimate for average real GDP growth over the next four quarters is about 2.94% a year. Half of the model's outcomes are above this and half below.

Simple exampleLine up all the outcomes the model thinks are possible for growth over the next year, from worst to best, and pick the one in the middle. That is this number: half of the outcomes are higher and half are lower. It is an estimate, not a promise.

Model noteCentral (median) model-implied path for average real GDP growth over the next four quarters.

Typical rangeBetween 2.72 and 2.99 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 2.53 to 3.11). The latest reading sits at the 69th percentile.

For the next quarter alone, the 1-in-20 bad case is growth of about -0.18% at an annual rate, meaning output would slightly shrink. Single quarters swing more than four-quarter averages, so this tail sits lower.

Simple exampleThis is the 1-in-20 bad case, but for just the next three months instead of a full year. A single quarter is like one game rather than a whole season: results swing more, so its bad case usually sits lower. A negative number means output would shrink in that bad case.

Model note5th-percentile model-implied annualized real GDP growth for the next quarter.

Typical rangeBetween -0.877 and -0.020 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -1.54 to 0.502). The latest reading sits at the 62th percentile.

Averaging across the worst 5% of outcomes for the next year gives growth of about 0.66% a year. Treat this as approximate: the fitted curve hit a limit of its settings (nu = 30), and its 5% point is 1.14 versus 1.17 from the regression.

Simple exampleRather than only asking where the worst 1 in 20 outcomes start, this averages all of them, like averaging your five worst test scores instead of noting the cutoff. It shows how deep the bad cases go, and it rests on a fitted curve, so treat it as approximate.

Model noteAverage growth in the worst 5% of model-implied outcomes over the next year; approximate, because the fitted skewed t is at its nu = 30 bound and misses the four quantile-regression quantiles by 0.060 pp RMSE (its 5% quantile is 1.14 vs 1.18 from the regression).

Typical rangeBetween -0.906 and 0.819 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -2.79 to 1.34). The latest reading sits at the 65th percentile.

When a measure of news and consumer sentiment is added to the model, the 1-in-20 bad case for next-year growth is about 1.51% a year. This version uses a shorter sample starting in 2002, so it rests on fewer observations.

Simple exampleThis is the same 1-in-20 bad case, but the model also listens to a measure of news and consumer mood, like asking how people feel in addition to checking the numbers. It uses a shorter history, so it rests on fewer examples.

Model noteSame 5% tail when the one-sided sentiment factor is added (2002-onward sample).

Typical rangeBetween 0.471 and 1.57 in half of all readings since Jan 2002 (middle 50%; the outer 10th to 90th percentile band is -0.894 to 2.16). The latest reading sits at the 68th percentile.

Adding sentiment improves the in-sample fit of the 5% bad-case line by 0.031 on a 0-to-1 scale, but random unrelated data improve it by a median of 0.010, and tests give p-values of 0.10 to 0.12. That means there is no statistically significant evidence that sentiment shifts the bad-case tail, and in a test on unseen data the gain was -0.050.

Simple exampleTo check whether a new clue really helps, compare it with adding random, meaningless data. If the new clue helps only a little more than random noise, as here, there is no solid evidence it adds anything. A test on data the model had not seen is the stricter check.

Model noteIn-sample improvement in the 5%-quantile fit from adding sentiment; it cannot be negative for nested models (irrelevant random regressors score a median 0.010), and against persistence-matched placebos and time-shifted sentiment its p = 0.10 to 0.12, i.e. no statistically significant evidence (5% level) that sentiment shifts the 5% tail (recursive out-of-sample quantile-score gain -0.050).

The model's most likely label is Goldilocks, meaning growth above its usual pace while inflation is slowing, with a probability of about 78.4%. The chance that growth is above trend is 83.5%, but this counts only part of the real uncertainty, so actual confidence should be lower.

Simple exampleThink of the economy's season, set by two questions: is growth faster than usual, and is inflation speeding up or slowing down? This shows which of the four combinations the model thinks is most likely, with its probability. A higher probability means more confidence, though the real uncertainty is larger than the model counts.

Model noteThe economy is classified as goldilocks with this model-implied probability (growth from the DFM activity factor, inflation from core PCE momentum); growth-side uncertainty is DFM signal-extraction only (P(growth above trend) = 84.6%), so it understates true growth uncertainty.

There is about a 78.4% model-implied chance that growth is above trend while core inflation is slowing. This is by far the largest of the four probabilities.

Simple exampleNamed after the porridge that was 'just right': growth faster than usual while price rises are slowing down. This is the model's estimated chance of that mix; higher means the model leans more strongly toward it.

Model noteProbability growth is above trend while core inflation is decelerating.

Typical rangeBetween 3.68 and 35.17 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 0.164 to 58.67). The latest reading sits at the 99th percentile.

There is about a 5.2% chance that growth is above trend while core inflation is speeding up. The low number mainly reflects that recent inflation momentum is clearly negative.

Simple exampleLike a busy shop that keeps raising its prices: growth faster than usual with inflation speeding up. This is the model's estimated chance of that combination; a higher number means it is more likely by the model's reckoning.

Model noteProbability growth is above trend while core inflation is accelerating.

Typical rangeBetween 3.47 and 43.06 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 0.353 to 68.42). The latest reading sits at the 31th percentile.

There is about a 1.0% chance of stagflation, meaning below-trend growth together with speeding-up inflation. Both of its conditions currently look unlikely to the model.

Simple exampleThe uncomfortable mix: the economy growing slower than usual while prices rise faster, like a town where fewer jobs are opening up but grocery bills climb quicker. This is the model's estimated chance of that combination.

Model noteProbability growth is below trend while core inflation is accelerating.

Typical rangeBetween 4.67 and 35.67 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 0.999 to 60.00). The latest reading sits at the 10th percentile.

There is about a 15.5% chance of a disinflationary slowdown, meaning below-trend growth with slowing inflation. It is the second most likely label, because inflation is slowing and growth could plausibly be below trend.

Simple exampleGrowth below its usual pace with inflation easing, like a party winding down: fewer people arriving and the music turned down. This is the model's estimated chance of that mix; a higher number means the model leans more toward it.

Model noteProbability growth is below trend while core inflation is decelerating.

Typical rangeBetween 4.67 and 45.77 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 0.667 to 68.88). The latest reading sits at the 44th percentile.

The growth reading is 0.26, the three-month average of the one-sided activity factor through August 2026. Positive means activity is growing faster than its average pace since 1990.

Simple exampleA speedometer for the economy, compared with its usual cruising speed. Above zero means activity is running faster than its long-run average; below zero means slower. The further from zero, the bigger the difference.

Model notePositive means activity growth is running above its post-1990 average.

Typical rangeBetween -0.251 and 0.217 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -0.585 to 0.447). The latest reading sits at the 79th percentile.

The inflation momentum reading is -0.96 percentage points: the 3-month annualized core PCE pace is that much below the 12-month pace. Negative means inflation is slowing.

Simple exampleThis compares how fast prices rose in the last three months with how fast they rose over the last year, like checking whether a runner's latest lap beat their average lap. Positive means inflation is picking up speed; negative means it is slowing. Prices can still be rising when it is negative, just more slowly.

Model noteNegative means recent core inflation is running below its 12-month pace (decelerating).

Typical rangeBetween -0.361 and 0.374 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -0.714 to 0.756). The latest reading sits at the 5th percentile.

The Goldilocks label has been the most likely one for 3 months in a row.

Simple exampleLike counting how many weeks a song has stayed at number one: this is how many months in a row the same season has been the model's top pick. A longer run means the picture has been steady.

Model noteHow long the current regime (argmax) has persisted.

This site's own US stress index is 0.037 on a scale from 0 (no stress) to 1 (extreme stress). The model describes this as low versus history.

Simple exampleThink of a smoke alarm system for the whole financial world. It reads 0 when everything is calm and climbs toward 1 when alarms are ringing in many places at once. A higher number means strain is wider and stronger.

Model noteSystemic stress in US markets is 0.037 on a 0-1 scale, low versus history.

Typical rangeBetween 0.028 and 0.124 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is 0.013 to 0.341). The latest reading sits at the 33th percentile.

The latest reading is at or above 35% of daily readings since 1986. In other words, about two thirds of past days were more stressed than today.

Simple exampleThis is like being told where your height ranks in your class. If a reading sits at the 80th percentile, it is more stressed than 80 out of every 100 past days. A low percentile means most past days were more stressed.

Model noteThe latest reading (as of 2026-10-15) is at or above 35% of daily readings since 1986.

Typical rangeThis reading is itself a percentile of its own history, so its middle 50% is 25 to 75 by construction. The latest reading is 35.

This shows how much extra stress comes from markets moving together. It nets out to about zero (+0.000): pairs of markets moving together add +0.028, while the government bond market moving against the others subtracts -0.027, so there is no net spreading effect right now.

Simple exampleOne alarm ringing may just be burnt toast, but alarms on every floor at once mean real trouble. This number is the extra worry added because markets are stressed together. If one market stays calm while others ring, it can pull the extra back down.

Model noteCorrelations net to about zero (+0.000): positively correlated segment pairs add +0.028 and negatively correlated pairs -0.027 (government bond market moving against the other segments), so aggregate stress shows no net systemic amplification.

Typical rangeBetween -0.010 and 0.062 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is -0.025 to 0.232). The latest reading sits at the 34th percentile.

The money market (short-term lending) accounts for 41% of today's stress reading, more than any other part. Its own score is 0.48, near the middle of the 0 to 1 range.

Simple examplePicture a team score broken down by player. The index adds up stress from five parts of the financial system, and this shows which part is contributing the biggest share of today's total, plus how stressed that part is on its own 0 to 1 scale.

Model noteMoney market (subindex 0.48) accounts for 41% of the composite stress reading.

The European Central Bank's own stress index for the United States is 0.018, at the 37th percentile of its history since 1990. It rose by 0.004 over the past month. It uses a different method, so its level is not directly comparable with this site's index.

Simple exampleThis is a second opinion, like seeing a different doctor who uses different tests. The European Central Bank builds its own US stress gauge from 0 (calm) to 1 (extreme). Because the recipe differs, compare whether the two move up and down together, not their exact levels.

Model noteThe ECB's United States CISS is 0.018, at the 37th percentile of its history since 1990 (+0.004 over one month).

Typical rangeBetween 0.014 and 0.129 in half of all readings since Jan 3, 2000 (middle 50%; the outer 10th to 90th percentile band is 0.006 to 0.265). The latest reading sits at the 31th percentile.

The European Central Bank's stress index for the euro area is 0.011, at the 16th percentile of its history since 1990. It fell by 0.023 over the past month.

Simple exampleThe same kind of stress thermometer, pointed at the countries that use the euro. It runs from 0 (calm) to 1 (extreme stress). The most useful comparison is with its own past readings: a higher percentile means more strain than usual for the euro area.

Model noteThe ECB's euro area CISS is 0.011, at the 16th percentile of its history since 1990 (-0.023 over one month).

Typical rangeBetween 0.023 and 0.231 in half of all readings since Jan 3, 2000 (middle 50%; the outer 10th to 90th percentile band is 0.009 to 0.403). The latest reading sits at the 13th percentile.

The European Central Bank's stress index for the United Kingdom is 0.003, at the 3rd percentile of its history since 1990, so it is near the calmest readings on record. It fell by 0.018 over the past month.

Simple exampleHere the stress thermometer is checking the United Kingdom. A reading near 0 means British markets are calm, and a reading near 1 means widespread strain. The percentile tells you how today compares with the UK's own history.

Model noteThe ECB's United Kingdom CISS is 0.003, at the 3rd percentile of its history since 1990 (-0.018 over one month).

Typical rangeBetween 0.013 and 0.135 in half of all readings since Jan 3, 2000 (middle 50%; the outer 10th to 90th percentile band is 0.005 to 0.331). The latest reading sits at the 5th percentile.

The European Central Bank's stress index for China is 0.001, below the 1st percentile of its history since 2006. It fell by 0.005 over the past month.

Simple exampleThis gauge checks financial stress in China, on the same 0 (calm) to 1 (extreme) scale. Its history is shorter than the others, so its percentile compares today only with the years it has been measured.

Model noteThe ECB's China CISS is 0.001, below the 1st percentile of its history since 2006 (-0.005 over one month).

Typical rangeBetween 0.019 and 0.100 in half of all readings since Oct 10, 2006 (middle 50%; the outer 10th to 90th percentile band is 0.008 to 0.201). The latest reading sits at the 0th percentile.

The index is 0.52 on a 0 to 1 scale, which the model labels moderate. All 7 of 7 indicators were available, so the reading uses the full set.

Simple exampleThink of a check-up with seven tests on how easily banks and big firms can borrow cash for a day or a few months. Each test gets a 0 to 1 score and they are averaged. Near 0 means borrowing is smooth; near 1 means it is unusually strained.

Model noteAverage rank of today's funding indicators versus their own history is 0.52 (moderate); 7 of 7 indicators available.

Typical rangeBetween 0.346 and 0.580 in half of all readings since Apr 6, 2018 (middle 50%; the outer 10th to 90th percentile band is 0.267 to 0.651). The latest reading sits at the 64th percentile.

The index is at the 66th percentile of its values since 2018-04-02, when all seven indicators first existed. That means it is higher than roughly two thirds of readings in that period.

Simple exampleThis ranks today against past readings, like a percentile on a school test. At the 70th percentile, today is more strained than 70 out of every 100 past days. A lower percentile means most past days looked more strained.

Model noteThe index sits at the 66th percentile of its values since all 7 indicators available at the as-of exist (2018-04-02).

Typical rangeThis reading is itself a percentile of its own history, so its middle 50% is 25 to 75 by construction. The latest reading is 66.

The overnight repo rate (SOFR) is 1 basis point below the rate the Fed pays banks on their reserves (IORB). Repo trading at or below that rate shows no sign of repo pressure.

Simple exampleBanks can earn a set interest rate just by leaving money at the Fed, like a safe savings account. If overnight loans backed by government bonds start costing more than that rate, cash is harder to find. At or below it means no sign of that squeeze.

Model noteSOFR is 1 bp below IORB (72nd percentile of its history), close to the administered rate and about 9 bp above its abundant-reserve norm (-10 bp); repo is firm though not above IORB.

Typical rangeBetween -10.00 and -1.00 in half of all readings since Apr 6, 2018 (middle 50%; the outer 10th to 90th percentile band is -10.00 to 3.00). The latest reading sits at the 72th percentile.

Across the day's repo trades, the highest rates were 12 basis points above the lowest (99th minus 1st percentile). A wider gap signals more uneven access to cash; this number has no fixed danger line here and enters the index as one of the seven ranked signals.

Simple examplePicture shoppers buying the same item. If everyone pays about the same price, the market is smooth. If some pay far more than others, access is uneven. This measures the gap between the highest and lowest rates paid for overnight cash on the same day.

Model noteThe 99th-1st percentile range of overnight repo rates is 12 bp (9th percentile of its history) - narrow; wide ranges signal uneven access to cash.

Typical rangeBetween 14.00 and 18.00 in half of all readings since Apr 6, 2018 (middle 50%; the outer 10th to 90th percentile band is 12.00 to 25.00). The latest reading sits at the 9th percentile.

On 2026-10-13, 3-month unsecured borrowing by highly rated financial firms (commercial paper) yielded about 1.08 basis points less than a 3-month Treasury bill, after putting both on the same yield basis. A value below zero means lenders asked for no extra yield over Treasury bills that day.

Simple exampleLending to a strong company for 3 months is a little riskier than lending to the US government. This is the extra interest the company has to pay. A bigger gap means lenders want more reward for that risk; zero or below means they are asking for no extra.

Model noteAA-rated financial issuers pay 1.1 bp less than Treasury bills for 3-month funding on a bond-equivalent basis (1st percentile of its history): no unsecured-funding premium over bills.

Typical rangeBetween 9.22 and 29.71 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is 6.09 to 48.00). The latest reading sits at the 1th percentile.

Bank reserves held at the Fed equal 11.6% of commercial-bank assets. Lower readings mean reserves are scarcer relative to the size of the banking system; this number is ranked against its own long history rather than compared with a fixed cutoff.

Simple exampleBank reserves are like the cash a household keeps in its checking account, measured against the size of its whole budget. A lower share means less spare cash relative to the size of the banking system, which can make short-term borrowing tighter.

Model noteReserves are 11.62% of commercial-bank assets (H.8 assets of 2026-09-30; 11.58% with the newer 2026-10-07 assets), below the roughly 12-13% at which Afonso et al. (2022) find the reserve demand curve starts to slope; lower readings mean scarcer reserves.

Typical rangeBetween 5.17 and 14.46 in half of all readings since Dec 20, 2002 (middle 50%; the outer 10th to 90th percentile band is 0.109 to 17.09). The latest reading sits at the 44th percentile.

Only about $0.1 billion sat in the Fed's overnight reverse repo facility on 2026-10-16. Near-zero use means this buffer of extra cash has been drained. The model notes that this signal reads as strained whenever the facility is empty, even when markets are calm, so a version of the index without it is also computed.

Simple exampleThe Fed runs a parking lot where spare cash can sit overnight. When the lot is full, there is plenty of extra cash around; when it is nearly empty, that cushion has been used up. An empty lot can happen in calm times too, so it is not a warning by itself.

Model noteON RRP take-up is $0.1bn, near zero: the Fed's reverse-repo buffer of excess liquidity has been drained.

Typical rangeBetween 4.90 and 356.9 in half of all readings since Sep 23, 2013 (middle 50%; the outer 10th to 90th percentile band is 0.548 to 1,771). The latest reading sits at the 6th percentile.

The model puts S&P 500 volatility at 11.9% a year, about 0.71 times its long-run level of 16.6%, so swings are smaller than usual. This reading uses returns up to the prior session; after that day's -1.10% move, the model's next-day estimate rises to 14.5% (0.87 times long-run).

Simple exampleVolatility is how bumpy the ride is for stock prices, like checking how choppy the sea is before a boat trip. This is the model's estimate of how big daily swings are likely to be, stated per year. Higher than the long-run level means a bumpier ride than usual.

Model noteModel-filtered S&P 500 volatility for 2026-10-16 is 11.9% annualized, 0.71x its GJR long-run level of 16.6%; this uses returns through the prior session, and after that day's -1.10% move the next-day model vol is 14.5% (0.87x).

Typical rangeBetween 9.56 and 18.23 in half of all readings since Oct 21, 2016 (middle 50%; the outer 10th to 90th percentile band is 8.13 to 25.79). The latest reading sits at the 47th percentile.

16.6% a year is the normal level the model expects S&P 500 volatility to drift back toward over time. It is estimated from about 10 years of data, so it reflects that period only.

Simple exampleAfter a storm or a calm spell, the sea tends to settle back to its usual choppiness. This is that usual level for stock-price swings, as the model sees it. It is learned from roughly the past decade, so it describes that period, not all of history.

Model noteUnconditional volatility implied by the fitted S&P 500 GJR-GARCH(1,1)-t parameters (in-sample, ~10 years of data).

The yen-dollar exchange rate is swinging at about 8.6% a year, compared with a long-run level of 10.9% (0.79 times), so currency moves are calmer than usual. Including that day's move, the next-day estimate is 8.4%.

Simple exampleIf you swap dollars for Japanese yen on a trip, the exchange rate moves a little every day. This measures how big those moves are, per year. A reading below the long-run level means the exchange rate is calmer than usual; above means it is swinging more.

Model noteYen-dollar volatility for 2026-10-09 is 8.6% annualized versus a long-run 10.9% (0.79x), using data through the prior observation; including that day's move the next-day model vol is 8.4%.

Typical rangeBetween 7.77 and 11.02 in half of all readings since Jan 4, 2000 (middle 50%; the outer 10th to 90th percentile band is 6.49 to 12.89). The latest reading sits at the 38th percentile.

The VIX was 16.5, which is 3.2 volatility points above the model's expected volatility for the next 22 trading days. The model reads this as option prices building in more volatility than the model expects, which it calls rich.

Simple exampleOption prices act a bit like insurance against big swings. This compares the swings that insurance prices build in (the VIX) with the swings the model expects. A positive gap means the insurance is priced for more turbulence than the model forecasts, which is the usual case.

Model noteVIX of 16.5 is 3.2 vol points above GJR-expected 22-day vol, so implied vol looks rich versus the model.

Typical rangeBetween 0.195 and 4.00 in half of all readings since Nov 18, 2016 (middle 50%; the outer 10th to 90th percentile band is -2.13 to 6.31). The latest reading sits at the 69th percentile.

Three common forces explain 68% of how 13 assets move together, at the 53rd percentile since 2018. Higher means markets are more tightly linked, so shocks spread more easily; today's reading is near the middle of that history.

Simple examplePicture a stadium crowd. Sometimes everyone does the wave together; other times people move on their own. This measures how much of the movement of 13 assets is explained by a few shared forces. Higher means they move more like one crowd, so a shock can spread more easily.

Model noteThe top 3 principal components explain 68% of cross-asset correlation variance (53rd percentile since 2018); higher means more tightly coupled markets.

Typical rangeBetween 0.655 and 0.682 in half of all readings since Dec 7, 2018 (middle 50%; the outer 10th to 90th percentile band is 0.634 to 0.688). The latest reading sits at the 53th percentile.

Market linkage over the latest 15 observations is 0.71 standard deviations above its average over the prior 252. The method's authors flag jumps above +1 standard deviation as signs of fragility, so this is above average but below that line.

Simple exampleThis compares the last few weeks of crowd togetherness with the usual level over the past year, measured in typical-sized wiggles. A reading near zero is normal. A big jump upward means markets have suddenly started moving much more in lockstep.

Model noteCoupling is 0.71 standard deviations above its trailing 252-observation average (about 1.1 years of common trading dates); KLPR flag spikes above +1 s.d. as fragility signals.

Typical rangeBetween -1.19 and 1.12 in half of all readings since Jan 17, 2020 (middle 50%; the outer 10th to 90th percentile band is -1.73 to 1.54). The latest reading sits at the 63th percentile.

Over the last 63 trading days, stocks and Treasury bonds have had a correlation of +0.41. Positive means they tend to fall together, which happens when inflation and interest rates drive markets; negative would mean bonds tend to gain when stocks drop.

Simple exampleCorrelation tells you whether two things tend to move the same way, like ice cream and sunscreen sales both rising in summer. Here it is stocks and government bonds. Positive means they tend to rise and fall together; negative means bonds tend to go up when stocks go down.

Model noteStocks and Treasuries have a +0.41 63-day correlation; positive means they sell off together (inflation/rates-driven regime), negative means bonds hedge.

Typical rangeBetween -0.393 and 0.143 in half of all readings since Jan 20, 2017 (middle 50%; the outer 10th to 90th percentile band is -0.515 to 0.349). The latest reading sits at the 94th percentile.

The ratio of 1-month to 3-month expected volatility is 0.975, below 1, so the curve has its normal upward shape (contango). It is at the 88th percentile since 2009 (89th over 5 years), so the near term is priced closer to the longer term than on most days.

Simple exampleThink of a weather forecast for the next month versus the next three months. Normally there is less expected turbulence in the near term than over the longer stretch, so this ratio sits below 1. If it rises above 1, traders are pricing more trouble soon than later.

Model noteThe 1-month to 3-month implied-vol ratio is 0.975 (normal; contango), at the 88th percentile of its history since 2009 and the 89th of the last 5 years.

Typical rangeBetween 0.834 and 0.930 in half of all readings since Sep 18, 2009 (middle 50%; the outer 10th to 90th percentile band is 0.802 to 0.979). The latest reading sits at the 89th percentile.

The 9-day VIX (20.1) is just above the 30-day VIX (20.0), a ratio of 1.007, at the 76th percentile since 2011 (78th over 5 years). This ratio has been above 1 on 26% of days since 2011, often around scheduled events, so this is a modest front-end premium and not by itself a stress signal.

Simple exampleThis compares expected stock-market swings over the next 9 days with the next 30 days. It often pops above 1 around known events, like a big announcement on the calendar, much as a forecast spikes for a day a storm is already scheduled.

Model noteThe very front end is inverted at 1.007 (9-day vol 20.1 vs 30-day 20.0), the 76th percentile since 2011 and 78th over 5 years; VIX9D/VIX is above 1 on 26% of days since 2011, often around scheduled event risk (FOMC, CPI, earnings), so this is a modest front-end premium, not by itself a stress signal.

Typical rangeBetween 0.847 and 0.973 in half of all readings since Jan 7, 2011 (middle 50%; the outer 10th to 90th percentile band is 0.801 to 1.04). The latest reading sits at the 84th percentile.

Option prices imply volatility of 20.7 points for the window from 30 to 93 days ahead, a bit above the current VIX of 20.0. That is near the middle of its history: the 55th percentile over the full sample and the 45th over 5 years.

Simple exampleIf you know how bumpy a whole 3-month trip is expected to be, and how bumpy the first month is, you can work out the bumpiness of the remaining stretch. This is that leftover window, from about 1 to 3 months ahead, as option prices see it.

Model noteThe market-implied volatility for the window from 30 to 93 days ahead is 20.7 vol points versus 20.0 spot VIX (above spot); percentile 55th full / 45th 5y.

Typical rangeBetween 16.79 and 23.91 in half of all readings since Sep 18, 2009 (middle 50%; the outer 10th to 90th percentile band is 15.47 to 29.42). The latest reading sits at the 55th percentile.

VVIX, which measures uncertainty about the VIX itself, is 109.9, or 5.50 times the VIX. It is at the 85th percentile since 2006 (82nd over 5 years), so traders are more unsure than usual about where volatility is heading.

Simple exampleThe VIX is a forecast of how stormy stocks will be. VVIX measures how unsure people are about that forecast itself, like weather forecasters disagreeing a lot. Higher means more uncertainty about where turbulence is heading.

Model noteUncertainty about volatility itself is 109.9 (VVIX/VIX 5.50), 85th percentile since 2006 and 82nd over 5 years.

Typical rangeBetween 82.57 and 101.8 in half of all readings since Mar 17, 2006 (middle 50%; the outer 10th to 90th percentile band is 75.44 to 115.5). The latest reading sits at the 85th percentile.

SKEW is 168.6, where 100 would mean no extra pricing of crash risk and higher means more is paid for crash protection. It is at the 99.7th percentile since 1990 and the 98th over 5 years, below the record of 183.1 on 2025-02-18.

Simple exampleSKEW is like the price of insurance against a sudden big drop in stocks, compared with ordinary insurance. 100 means no extra charge for crash protection; higher means people are paying more for it. It shows demand for protection, not a forecast that a crash will happen.

Model noteSKEW is 168.6 (risk-neutral skewness S = -6.86; 100 = symmetric, higher = more priced crash protection), 99.7th percentile since 1990 and 98th over the last 5 years (record 183.1 on 2025-02-18).

Typical rangeBetween 117.1 and 133.6 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is 112.8 to 145.5). The latest reading sits at the 100th percentile.

Over the last 5 years, SKEW has been at or below today's level about 98% of the time. The full-history rank (99.7th) looks even higher partly because SKEW has drifted upward since 2010.

Simple exampleRanking a number against its past is like comparing your height with past classes. If kids have been getting taller over the years, you look taller next to your grandparents' class than next to recent ones. SKEW has drifted up over time, so its long-history rank looks higher than its recent rank.

Model noteSKEW ranks at the 98th percentile of the trailing 5 years and the 99.7th of the full 1990-on history; the full-history rank is higher than the 5-year rank, which is consistent with the upward drift of SKEW since 2010 (full-history ranks overstate extremity).

Typical rangeThis reading is itself a percentile of its own history, so its middle 50% is 25 to 75 by construction. The latest reading is 98.

The 1-month versus 3-month curve is not inverted today (0 days in a row). It was last inverted on 2026-04-07 and was inverted on 13 of the last 252 trading days.

Simple exampleThis is a streak counter, like counting rainy days in a row. It counts how many trading days in a row near-term expected turbulence has been priced above longer-term turbulence, and how often that happened in the past year.

Model noteThe curve is not inverted today; last inverted 2026-04-07 (13 of the last 252 days).

Weeks with at least one inverted day had an average systemic stress reading (CISS) of 0.105, versus 0.061 in other weeks, a gap of 0.044. The gap is not statistically significant at the 5% level (t = 1.90), and only 26% of the highest-stress weeks were inverted, so the link is weak; part of it is built in, because the VIX is also an input to the CISS.

Simple exampleThis checks whether wider financial stress tends to be higher in weeks when near-term turbulence was priced above longer-term turbulence, a bit like checking whether days when people carry umbrellas really are rainier. Part of the link is built in, because both measures share an ingredient.

Model noteWeeks containing an inverted day had mean CISS 0.105 versus 0.061 otherwise (HAC t = 1.90 with 8 Newey-West lags, not significant at 5%; t = 1.71/1.66 with 26/52 lags), and only 26% of the top-decile-stress weeks were inverted, so the validation is weak (and partly mechanical: VIX is a CISS input).

The best-fitting model puts a 3% probability on the Nasdaq being in its high-volatility state this week, so it sees the market as very likely in its calm state. The estimate uses data up to this week, with model settings fitted on the full history.

Simple exampleLike guessing whether it is storm season from the last few weeks of weather. This is the model's estimated chance that the Nasdaq stock index is in its stormy, big-swing state rather than its calm one. Near 100% means the model is fairly sure it is stormy; near 0% means it likely is not. It is an estimate, not a certainty.

Model noteThe M0 Markov-switching model puts a 3% probability on Nasdaq being in its high-volatility regime this week (one-sided filter given full-sample parameters).

Typical rangeBetween 0.019 and 0.617 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is 0.014 to 0.953). The latest reading sits at the 37th percentile.

In the stress state, weekly Nasdaq returns swing at about 35% a year. This describes how turbulent the stormy state is on average, not today's level.

Simple exampleThink of the typical size of waves on a rough day at sea. This is how big the Nasdaq's yearly price swings usually are when the model's stormy state is in charge. It describes that state on average, not what is happening today.

Model noteIn the stress regime weekly Nasdaq returns have an annualized volatility of about 35%.

In the calm state, swings are about 14% a year, about 2.4 times smaller than in the stress state.

Simple exampleThe same wave-size measure, but for a calm day at sea. It shows how big Nasdaq swings typically are in the quiet state; the bigger its gap from the stormy number, the more different the two states are.

Model noteIn the calm regime annualized volatility is about 14%, 2.4x lower than in stress.

With today's switching odds, a calm spell lasts about 61 weeks on average. This is an average implied by the model's switching probability, not a forecast of when the calm ends.

Simple exampleIf a sunny spell has a small chance of ending each week, you can work out how long sunny spells last on average. This reading does that for the market's current state. It is an average, like an average bus wait, not a prediction of when the state will change.

Model noteWith transition probabilities at today's values, a calm spell lasts 61 weeks on average (1/(1-p_ii)).

In the version that adds news mood (M1), better news mood lowers the odds of switching from calm to stress: the coefficient is -1.12 (standard error 0.41, p = 0.006), significant at the 5% level. When the stress index is also added (M2), the effect is smaller at -0.80 (p = 0.014).

Simple exampleThis asks: does cheerful news make a calm market less likely to turn stormy? A negative number means better news mood goes with lower odds of switching to stress, and the further from zero, the stronger the link. A small p-value means a link this strong would be unlikely to appear by chance if there were no real effect.

Model noteBetter news mood lowers the odds of switching from calm to stress (coef -1.12, SE 0.41, p = 0.006: significant at 5%); in M2, with CISS added, -0.80 (SE 0.32, p = 0.014).

Adding news mood raises the model's fit score (log-likelihood) by 4.6, giving a likelihood-ratio statistic of 9.2 and p = 0.010. On this test alone, news mood improves the fit.

Simple exampleLike testing whether a new ingredient really improves a recipe. This compares the model with and without news mood: a small p-value means the version with news mood fits the data better than luck alone would easily explain.

Model noteAdding news mood to both transition equations raises the log-likelihood by 4.6 (LR = 9.2, p = 0.010).

The BIC score, which penalizes extra parameters, prefers the simplest model M0 (constant switching odds) by 4.7 points: 9215.8 versus 9221.7 for M1 and 9220.5 for M2, where lower is better. So the evidence on news mood is mixed: the likelihood-ratio test favors it, but the stricter BIC does not.

Simple exampleBIC is a judge that rewards a good fit but takes off points for every extra moving part, so simpler models get a fair chance. Lower scores win. This reading shows which version wins and by how many points; a bigger margin means a clearer winner.

Model noteBIC prefers M0 by 4.7 points; BIC: M0 9215.8, M1 9221.7, M2 9220.5.

The 5-year rate built into bond prices for the period 5 to 10 years from now is 5.52% (5.67% with annual compounding). It includes term premia, the extra pay for lending for longer, so it is not a pure expectation of future rates.

Simple examplePicture agreeing today on the rent for an apartment you will move into in 5 years and keep for 5 years. This is the yearly interest rate that today's bond prices lock in for that future stretch. It also includes extra pay for tying money up longer, so it is not simply a guess about future rates.

Model noteThe 5-year rate implied for the period five to ten years ahead is 5.52% (continuous compounding, 5.67% annually compounded); it includes term premia, not pure expectations.

Typical rangeBetween 2.96 and 5.24 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is 2.26 to 5.94). The latest reading sits at the 82th percentile.

The 1-year rate for the year starting 2 years from now is 5.13%, above today's 1-year zero yield of 4.49%. Because it includes term premia, it is not the expected path of the 1-year rate.

Simple exampleLike booking today the price of a one-year rental that starts two years from now. This is the interest rate bond prices imply for that single future year. Because it includes some extra pay for lending longer, it is not a pure forecast of where rates will be.

Model noteThe 1-year forward rate for the year starting two years ahead is 5.13%, above today's 1-year zero yield of 4.49%; it includes term premia, so it is not the expected path of the 1-year rate.

Typical rangeBetween 1.65 and 3.95 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is 0.795 to 4.68). The latest reading sits at the 95th percentile.

The 10-year zero yield is 49 basis points above the 2-year, so the curve slopes upward. On the standard par Treasury curve the same gap is 0.44 percentage points.

Simple exampleThink of a staircase of interest rates by loan length. This is the 10-year rate minus the 2-year rate: positive means longer loans pay more, so the stairs go up; negative means short loans pay more. It is shown in basis points, hundredths of a percentage point.

Model noteThe zero-coupon curve is upward sloping by 49 bp between 2 and 10 years (par CMT 10y-2y: 0.44 pp on 2026-10-09).

Typical rangeBetween 0.270 and 2.19 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is -0.055 to 2.70). The latest reading sits at the 32th percentile.

The fitted long-run level of zero yields is 5.87%, compared with 5.65% on par Treasury yields. The same fit gives a slope of 0.92 percentage points and a curvature of -2.67, with an average fit error of 8.8 basis points.

Simple exampleThink of the overall height of a hill rather than its shape. This summary number is roughly where interest rates settle for very long loans. Higher means rates for long loan lengths are higher overall.

Model noteDiebold-Li level on 1-30y zeros is 5.87% versus 5.65% on par CMT yields (slope 0.92 pp, curvature -2.67 pp, fit RMSE 8.8 bp).

Typical rangeBetween 3.32 and 5.36 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is 2.67 to 6.06). The latest reading sits at the 85th percentile.

The fitted gap between very long and very short zero yields is 0.92 percentage points, compared with 1.49 on par yields. A positive value means long-term rates sit above short-term rates.

Simple exampleThis tells you how much the hill tilts: the gap between very long-term and very short-term rates. Positive means long loans pay more than short ones; negative means short loans pay more.

Model noteThe fitted long-minus-short zero slope is 0.92 pp versus 1.49 pp on par yields.

Typical rangeBetween -0.001 and 3.06 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is -0.773 to 4.41). The latest reading sits at the 41th percentile.

The 5y5y forward rate rose by 39 basis points (hundredths of a percentage point) since 2026-09-09, about one month. That is the change in the rate built in for the period 5 to 10 years ahead.

Simple exampleLike checking how much a locked-in future rent price moved over the past month. This is how much the 5-year rate starting 5 years from now has changed in about a month, in basis points (hundredths of a percentage point). Positive means it rose.

Model noteThe 5y5y forward moved +39 bp since 2026-09-09.

Typical rangeBetween -18.39 and 16.44 in half of all readings since Jan 3, 2000 (middle 50%; the outer 10th to 90th percentile band is -33.05 to 36.31). The latest reading sits at the 91th percentile.

Rebuilding the curve from the Fed's published formula settings matches the published yields to within 0.038 basis points on all 16295 dates checked. This is a data-quality check confirming that the curve used here is computed correctly.

Simple exampleA quality check, like redoing a math problem to confirm you get the textbook's answer. The model rebuilds the Fed's published yields from its formula and measures the biggest mismatch. A tiny number means the rebuilt curve matches closely.

Model noteRe-building the curve from the published parameters matches SVENY01-30 to within 0.038 bp on every one of 16295 dates (100.00% below 1 bp).

One US dollar is worth 158.34 yen, 2.78% more than 21 trading days earlier. A rising number means a weaker yen, which the output notes works in favour of the carry trade.

Simple exampleThis is a price tag: how many yen it takes to buy one US dollar, a bit like how many arcade tokens you get per dollar. When the number goes up, each yen buys less, so the yen has weakened. When it goes down, the yen has strengthened.

Model noteOne dollar buys 158.34 yen, +2.78% versus 21 trading days earlier (positive = weaker yen, which favours the carry trade).

Typical rangeBetween 104.2 and 120.7 in half of all readings since Jan 3, 2000 (middle 50%; the outer 10th to 90th percentile band is 89.22 to 146.4). The latest reading sits at the 98th percentile.

Two-year US Treasuries pay 2.86 percentage points more per year than two-year Japanese bonds. The short-term gap (3-month US bills versus 1-year JGBs) was 2.51 points on 2026-10-09; a wider gap means a bigger interest reward for borrowing in yen.

Simple exampleThink of two banks, one paying a little interest and one paying more. This is the difference between what a 2-year US government bond pays each year and what a 2-year Japanese one pays. A wider gap means a bigger reward for borrowing where money is cheap and putting it where it earns more.

Model noteTwo-year Treasuries yield 2.86 pp more than 2-year JGBs; the short-rate carry (3m bill BEY minus 1y JGB) was 2.51 pp on 2026-10-09.

Typical rangeBetween 0.575 and 3.44 in half of all readings since Jan 4, 2000 (middle 50%; the outer 10th to 90th percentile band is 0.255 to 4.30). The latest reading sits at the 67th percentile.

The rate gap is 0.34 times the model's expected yearly swing in the dollar-yen rate (8.6%), so the reward covers about a third of one typical year's currency risk. That ranks at the 72nd percentile of its history since 2000, meaning it is higher than on about 72% of past days; higher readings mean the trade looks more attractive and possibly more crowded.

Simple examplePicture a lemonade stand with a small profit, where one bad-weather day can wipe it out. This reading compares the interest reward with how much the dollar-yen rate usually jumps around in a year. Higher means more reward for each unit of currency risk; lower means normal wobbles could easily swallow the reward.

Model noteThe 2y differential is 0.34x the model-expected annualized USD/JPY vol (8.6%), a percentile rank of 72 (of 100) in its history since 2000 (with 1m realized vol: 0.50); higher means a more attractive, more crowded-looking carry.

Typical rangeBetween 0.059 and 0.360 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is 0.028 to 0.488). The latest reading sits at the 72th percentile.

Over the last 21 trading days, a trade that borrowed yen to own dollars would have earned +2.94%, counting both the currency move and the interest gap. For scale, the worst 21-day stretch since 2000 lost 13.5%, from 2008-09-24 to 2008-10-24.

Simple exampleA monthly report card for the trade of borrowing yen to hold dollars. It adds up what the exchange-rate move and the interest gap would have earned or lost over roughly the last month of trading days. Positive means the trade made money over that stretch; negative means it lost.

Model noteOver the last 21 trading days the yen carry trade earned +2.94% (spot plus interest accrual); the worst 21-day loss since 2000 was 2008-09-24..2008-10-24 (-13.5%).

Typical rangeBetween -1.42 and 1.96 in half of all readings since Feb 4, 2000 (middle 50%; the outer 10th to 90th percentile band is -3.06 to 3.80). The latest reading sits at the 84th percentile.

The running carry index is 2.9% below its highest point since 2000, and also 2.9% below its peak of the past year. A drawdown near zero means the trade is close to its best level; a deep drawdown means recent losses.

Simple exampleThink of a hiker checking how far below the highest point of the trip they are right now. This shows how far the trade's running total sits below its best level so far. Zero means it is at the peak; a bigger number means it has slid further down.

Model noteThe carry index is 2.9% below its all-time (since 2000) peak and 2.9% below its 1-year peak.

Typical rangeBetween -16.40 and -3.50 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is -26.11 to -0.746). The latest reading sits at the 77th percentile.

Over the last 63 days, daily carry returns had a correlation of -0.40 with the S&P 500, meaning they tended to move in opposite directions. The output reads this as the August-2024 pattern (yen unwinds and stock falls happening together) not being active now, while noting that daily figures are noisy and a timing-robust version is -0.19.

Simple exampleWatch two dancers: do they step the same way or in opposite directions? This measures whether the carry trade's daily gains and losses have tended to move with the US stock market or against it. Near +1 means together, near -1 means opposite, and near zero means little link.

Model noteDaily carry returns have a -0.40 63-day correlation with the S&P 500 (-0.42 with Nasdaq, +0.41 with VIX changes); this is negative: carry returns have recently moved opposite to equities, so the August-2024 channel (a yen unwind hitting equities at the same time) is not currently active; because FX is the noon New York rate while equities are closing prices, daily correlations are noisy, and the timing-robust version on 63 non-overlapping 2-day sums is -0.19.

Typical rangeBetween -0.065 and 0.307 in half of all readings since Jan 20, 2017 (middle 50%; the outer 10th to 90th percentile band is -0.235 to 0.441). The latest reading sits at the 2th percentile.

The fragility score is -0.89, and only 6% of days since 2001-03-30 scored at or below it, so by this measure strain is low compared with history. The output warns that the score mostly rises during an unwind rather than before one; only its carry-to-risk part looks ahead, and that part sits at the 80th percentile.

Simple exampleA stress gauge that blends three warning signs: how tempting the trade looks, how much it recently lost, and how tied it is to stocks. This ranks today's score against all past days. A high rank means more strain than usual, though the gauge tends to rise once trouble starts rather than before.

Model noteThe heuristic fragility score is -0.89 (mean of expanding z-scores of carry-to-risk, 21-day drawdown depth and Nasdaq coupling); 6% of days since 2001-03-30 scored at or below today (higher = more fragile or stressed); it is largely a coincident gauge because the drawdown and coupling components jump during an unwind, and only the carry-to-risk component is ex-ante (its z-score has a percentile rank of 80 of 100).

Typical rangeThis reading is itself a percentile of its own history, so its middle 50% is 25 to 75 by construction. The latest reading is 6.

Japan's 10-year government bond yields 2.95%, and the gap between its 10-year and 2-year yields is 1.03 points. The US 10-year yield is 2.15 points higher than Japan's 10-year.

Simple exampleThis is the yearly interest Japan's government pays when it borrows for 10 years, like the rate on a long loan. Higher means long-term borrowing in yen costs more. The reading also shows how much more a 10-year US government loan pays than Japan's, and how Japan's 10-year rate compares with its 2-year rate.

Model noteThe 10-year JGB yields 2.95% with a 10y-2y slope of 1.03 pp; US 10y minus JGB 10y is 2.15 pp.

Typical rangeBetween 0.257 and 1.44 in half of all readings since Jan 4, 2000 (middle 50%; the outer 10th to 90th percentile band is 0.026 to 1.71). The latest reading sits at the 100th percentile.

Leveraged funds are net short yen futures equal to 9.0% of all open contracts, which is a bet on a weaker yen (the same direction as the carry trade). Its percentile rank over three years is 70, and with a z-score inside plus or minus 2 the model does not flag it as crowded.

Simple exampleBig leveraged funds, mostly hedge funds, can bet on the Japanese yen going up or down. This number takes their 'up' bets, subtracts their 'down' bets, and compares the difference with the size of the whole market. Below zero means they lean toward a weaker yen; above zero, a stronger one.

Model noteLeveraged funds are net short 9.0% of open interest (net short yen futures = short JPY = long USDJPY carry), 3-year percentile rank 70 (0-100), which is not extreme versus its 3-year history (|z| <= 2).

Typical rangeBetween -24.26 and 6.78 in half of all readings since Jan 8, 2008 (middle 50%; the outer 10th to 90th percentile band is -33.22 to 24.53). The latest reading sits at the 55th percentile.

The yen position is +0.47 standard deviations from its 3-year average, a fairly ordinary reading. It moved 13.3 percentage points of open interest toward short over the last four weeks; changes around quarterly contract rolls can partly reflect swings in total open interest.

Simple exampleLike comparing today's temperature with the usual range for your town, this shows how unusual the funds' yen bet is compared with the past three years. Near zero is ordinary, while readings beyond plus or minus 2 are rare, and the model then calls the market crowded.

Model notePositioning is +0.47 standard deviations from its trailing 3-year mean (-13.3pp over 4 weeks, report released 2026-10-16); for comparison, leveraged funds in euro futures are net short 7.8% of open interest with z -2.18 (report of 2026-10-13), a crowded short (z < -2 with a net short position), although the position itself is small (7.8% of open interest).

Typical rangeBetween -0.827 and 0.745 in half of all readings since Jun 3, 2008 (middle 50%; the outer 10th to 90th percentile band is -1.28 to 1.55). The latest reading sits at the 68th percentile.

Leveraged funds are net short 21.2% of S&P 500 e-mini open interest, at the 14th percentile of the last three years. These funds are usually net short here because of hedged trades against asset managers, so the level is read through its z-score; it is not flagged as crowded.

Simple exampleThe same 'up bets minus down bets' measure, now for futures on the S&P 500, a basket of large US companies. These funds usually lean toward 'down' here because they often use these contracts as one half of a paired trade, so a negative number is normal by itself.

Model noteLeveraged funds are net short 21.2% of open interest (against asset managers' net +47.4%: leveraged funds are structurally short e-minis via basis trades, so read it through its z-score, not its sign), 3-year percentile rank 14 (0-100), which is not extreme versus its 3-year history (|z| <= 2).

Typical rangeBetween -14.85 and -7.17 in half of all readings since Jan 8, 2008 (middle 50%; the outer 10th to 90th percentile band is -18.86 to -3.59). The latest reading sits at the 4th percentile.

The S&P 500 futures position is -1.23 standard deviations from its 3-year average, more short than usual but inside the crowding line of 2. It moved 9.2 points further short over four weeks.

Simple exampleBecause these funds usually lean the same way in stock futures, the useful question is whether they lean more or less than usual. This score answers that: zero is their typical lean over three years, and a big move away from zero in either direction means the position is unusually lopsided.

Model notePositioning is -1.23 standard deviations from its trailing 3-year mean (-9.2pp over 4 weeks, report released 2026-10-16).

Typical rangeBetween -0.750 and 0.636 in half of all readings since Jun 3, 2008 (middle 50%; the outer 10th to 90th percentile band is -1.41 to 1.56). The latest reading sits at the 14th percentile.

Leveraged funds are net short 37.9% of 10-year Treasury futures open interest, at the 42nd percentile of the past three years. That is close to the middle of its recent range and not flagged as crowded.

Simple exampleSame measure again, for futures on 10-year US government bonds, the IOUs the US government sells to borrow money. A more negative number means these funds hold more positions that gain when bond prices fall, often as one half of a paired trade rather than a simple bet.

Model noteLeveraged funds are net short 37.9% of open interest (against asset managers' net +49.8%: largely the cash-futures basis trade, so read it through its z-score, not its sign), 3-year percentile rank 42 (0-100), which is not extreme versus its 3-year history (|z| <= 2).

Typical rangeBetween -21.58 and 1.58 in half of all readings since Jan 8, 2008 (middle 50%; the outer 10th to 90th percentile band is -34.01 to 12.70). The latest reading sits at the 7th percentile.

The Treasury futures position is -0.35 standard deviations from its 3-year average, a small gap. It moved 3.2 points further short over four weeks.

Simple exampleThink of a report card that compares this term's grade with your own last three years. This score shows whether funds' bond-futures position is about normal for them (near zero) or unusually stretched (far from zero, especially beyond plus or minus 2).

Model notePositioning is -0.35 standard deviations from its trailing 3-year mean (-3.2pp over 4 weeks, report released 2026-10-16).

Typical rangeBetween -1.47 and 0.585 in half of all readings since Jun 3, 2008 (middle 50%; the outer 10th to 90th percentile band is -1.84 to 1.66). The latest reading sits at the 57th percentile.

Over the 13 weeks to 2026-10-10, large Japanese investors sold JPY 3.24 trillion more foreign bonds than they bought. The 3-year z-score of -1.18 says this outflow is larger than typical but not extreme.

Simple examplePicture large Japanese investors as shoppers for other countries' bonds. This adds up what they bought minus what they sold over the last 13 weeks. A positive total means they bought more foreign bonds than they sold; a negative total means they sold more than they bought.

Model noteDesignated Japanese investors sold a net JPY 3.24tn of foreign bonds over 13 weeks (3y z -1.18; week ending 2026-10-10, released 2026-10-16).

Typical rangeBetween -0.808 and 5.01 in half of all readings since Mar 29, 2014 (middle 50%; the outer 10th to 90th percentile band is -4.24 to 7.42). The latest reading sits at the 12th percentile.

Japan owns about $1068 billion of US Treasuries, which is 11.4% of all foreign-owned Treasuries. That stock is $116 billion lower than 12 months earlier (end-August 2026 data).

Simple exampleWhen the US government borrows, it sells IOUs called Treasuries, and Japan is one of the biggest foreign holders of them. This shows how much Japan holds, a bit like checking how much one big friend has lent you. A rising number means Japan holds more; a falling one means less.

Model noteJapan held $1068bn of Treasuries at end-Aug 2026 (11.4% of foreign holdings), a change in holdings of -116bn over 12 months that includes price/valuation effects and is not the same as net purchases (released 2026-10-16).

Typical rangeBetween 1,102 and 1,215 in half of all readings since Dec 2011 (middle 50%; the outer 10th to 90th percentile band is 1,067 to 1,266). The latest reading sits at the 10th percentile.

Global growth momentum is +0.26 standard deviations above its average, and it changed by only -0.08 over three months, which the model calls broadly stable. The OECD's GDP-weighted G20 indicator fell 0.17 points over six months, the opposite sign, because this factor weights countries by their shared pattern, not by economic size.

Simple exampleThink of 11 economies as runners in one group. This number tracks whether the group as a whole is moving faster or slower than its usual pace: above zero means faster than normal, below zero means slower. Countries that move with the pack count more here than the biggest ones do.

Model noteGlobal growth momentum is above average (+0.26 sd) and broadly stable over three months (-0.08 sd; moves within +-0.10 sd are called broadly stable). The OECD's GDP-weighted G20 CLI fell 0.17 points over six months, the opposite sign: the factor weights the standardized economies by PCA loadings, not by GDP.

Typical rangeBetween -0.500 and 0.544 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -1.34 to 1.12). The latest reading sits at the 61th percentile.

73% of the 11 economies have a leading indicator higher than three months earlier. A share above half means more economies are improving than weakening.

Simple exampleLike asking a class whether each student's grades went up since last term, this is the share of countries answering yes for their economic outlook. More than half means more economies are improving than slipping; less than half means more are slipping.

Model note73% of the 11 economies with a published 2026-09 OECD leading indicator have a higher reading than three months earlier.

Typical rangeBetween 27.27 and 81.82 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 9.09 to 90.91). The latest reading sits at the 64th percentile.

91% of the 11 economies are above their long-term trend. By phase, 8 are in expansion, 2 in downturn and 1 in contraction.

Simple exampleEvery economy has a normal cruising speed. This is the share of countries currently running faster than their own cruising speed. A high share means most economies are above their usual pace, even if some of them are starting to slow down.

Model note91% of the 11 economies have an above-trend leading indicator; phases (each economy's latest published month): contraction 1, downturn 2, expansion 8.

Typical rangeBetween 18.18 and 81.82 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 9.09 to 90.91). The latest reading sits at the 80th percentile.

The average correlation between countries' 6-month indicator changes is 0.07, at the 6th percentile of its history since 1992 (median 0.48). In plain terms, national economies are moving much less in step than usual.

Simple examplePicture a choir: are the singers going up and down together, or each doing their own thing? A high number means countries' economies rise and fall in step. A low number means they are following separate paths, so good or bad news in one place is less often shared by the others.

Model noteThe average pairwise correlation of 6-month CLI changes is 0.07, at the 6th percentile of its history since 1992 (median 0.48): national cycles are currently unusually unsynchronized.

Typical rangeBetween 0.425 and 0.712 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 0.172 to 0.776). The latest reading sits at the 0th percentile.

44% of the 9 tracked central banks raised interest rates over the last six months. The median change across all of them was +0 basis points.

Simple exampleCentral banks set the basic price of borrowing money, a bit like setting the rent on a loan. This is the share of tracked central banks that raised that price over the last six months. A higher share means borrowing is getting more expensive in more countries.

Model note44% of the tracked central banks raised rates in the past six months (median change +0 bp).

Typical rangeBetween 11.11 and 44.44 in half of all readings since Feb 2001 (middle 50%; the outer 10th to 90th percentile band is 0.000 to 64.44). The latest reading sits at the 67th percentile.

11% of the 9 tracked central banks cut interest rates over the last six months, so more banks have been raising than cutting.

Simple exampleThe flip side of the hiking count: the share of tracked central banks that lowered their main interest rate over the last six months. When many are cutting, borrowing for homes and businesses tends to get cheaper in more countries.

Model note11% of the tracked central banks cut rates in the past six months.

Typical rangeBetween 11.11 and 44.44 in half of all readings since Feb 2001 (middle 50%; the outer 10th to 90th percentile band is 0.000 to 66.67). The latest reading sits at the 17th percentile.

The New York Fed's supply-chain pressure index is +1.28 standard deviations above its historical average and rose 0.08 over three months. Positive readings mean shipping and supply conditions are tighter than normal.

Simple exampleA pizza that usually arrives in 30 minutes shows up late and costs more when roads are jammed and ingredients are scarce. This index measures that kind of strain for global shipping and supplies: above zero means tighter than normal, below zero means smoother than normal.

Model noteSupply-chain pressure is +1.28 sd from average and changed +0.08 over three months.

Typical rangeBetween -0.609 and 0.269 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -0.818 to 1.28). The latest reading sits at the 90th percentile.

None of the 9 economies has a credit-to-GDP gap above 10 points, the zone where Basel rules point to the maximum bank capital buffer. Two more economies are in the middle band of 2 to 10 points (data for quarters ending 2026-03-31).

Simple exampleIf a family's borrowing grows much faster than its income, trouble can follow later. This counts countries whose private borrowing, compared with the size of their economy, is far above its usual trend, the zone where international bank guidelines point to the biggest safety cushion for banks.

Model note0 economies are in the Basel maximum-buffer zone (gap above 10pp) and 2 more are in the 2-10pp band using the latest published BIS gaps (quarters ending 2026-03-31).

The combined mood factor is -0.97 standard deviations, below average, based on 3 of its 5 inputs this month. It sits at the 18th percentile of the full 1997-2026 history but the 44th of the one-sided history, which drifts downward over time (its average in expansion months is -0.73), so the change matters more than the level.

Simple exampleThink of a class mood score built from several surveys at once: news tone, how shoppers feel, and how worried people are about policy and world events. Above zero means a brighter mood than usual, below zero a gloomier one. Because this score drifts lower over the years, its changes say more than its exact level.

Model notePC1 of news, consumer, policy-uncertainty and geopolitical mood is -0.97 sd (higher = better mood; 3 of 5 inputs this month), the 18th percentile of the 1997-2026 full-sample factor but the 44th of the one-sided history, which drifts down (one-sided expansion-month mean -0.73).

Typical rangeBetween -1.48 and -0.095 in half of all readings since Dec 2001 (middle 50%; the outer 10th to 90th percentile band is -1.97 to 0.305). The latest reading sits at the 43th percentile.

The news mood is -0.15, a slightly below-normal reading compared with each source's past three years. It is at the 39th percentile of business days since 1990.

Simple exampleA weather reading for the headlines: are news stories sounding cheerful or worried compared with the past few years? A higher number means more upbeat coverage; a lower number means gloomier coverage than normal.

Model noteThe average 3-year rolling z-score of the news-based mood components is -0.15 (39th percentile of business days since 1990).

Typical rangeBetween -0.552 and 0.490 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is -1.11 to 0.886). The latest reading sits at the 43th percentile.

The market-implied mood is +0.74, meaning low volatility, narrow credit spreads and easy financial conditions compared with the past three years. It is at the 73rd percentile of business days since 1990.

Simple exampleInstead of asking people, this reads the mood from markets, like judging a crowd's nerves by how jumpy it is. Calm stock prices, little extra interest demanded on company loans and easy borrowing push it higher. A lower reading means markets are acting more nervous than usual.

Model noteThe average 3-year rolling z-score of -VIX, -Baa spread and -NFCI is +0.74 (73rd percentile of business days since 1990).

Typical rangeBetween -0.314 and 0.826 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is -1.35 to 1.18). The latest reading sits at the 71th percentile.

The gap between market mood and news mood is about 0.9, positive, so markets are calmer than the news flow. The output labels this the complacency side.

Simple examplePicture the forecast calling for a storm while everyone walks around in shorts. This is market mood minus news mood: positive means markets are calmer than the headlines, negative means markets are more nervous than the news suggests.

Model noteMarkets are calmer than the news flow (complacency side).

Typical rangeBetween -0.337 and 0.839 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is -1.06 to 1.31). The latest reading sits at the 78th percentile.

The gap is at or above its level on 80% of business days since 1990. So markets being calmer than the news is more pronounced than usual.

Simple exampleThis ranks today's gap against every past business day, like saying you are taller than 90 out of 100 classmates. A high rank means markets are unusually calm compared with the news; a low rank means unusually nervous compared with it.

Model noteThe gap is at or above its level on 80% of business days since 1990.

Typical rangeThis reading is itself a percentile of its own history, so its middle 50% is 25 to 75 by construction. The latest reading is 80.

The latest Federal Reserve (FOMC) statement scored +1, with 2 hawkish phrases and 0 dovish ones. Because recent statements are short, a few phrases can push the score to its maximum of +1.

Simple examplePicture counting words in a coach's speech: 'we need to tighten up' versus 'let's ease off'. This score counts phrases in the US Federal Reserve's statement that point to rising prices or strong activity (hawkish) against those pointing to weakness (dovish). It runs from -1, all dovish, to +1, all hawkish.

Model noteLatest FOMC statement has 2 hawkish vs 0 dovish phrases (+1 = all hawkish).

The FOMC tone was unchanged (+0.00) compared with the 2026-07-29 statement.

Simple exampleThis compares the latest Federal Reserve statement's tone with the one before it, like noticing whether a teacher's comments got stricter or softer since last term. Positive means more hawkish wording, negative means more dovish, and zero means no change.

Model noteTone moved +0.00 vs the 2026-07-29 statement (unchanged).

The latest scored Bank of Japan statement had 16 hawkish phrases and 0 dovish ones, a tone of +1. That means its wording leaned fully toward strength in prices or activity.

Simple exampleThe same word-counting score, applied to the Bank of Japan's statements. Near +1 means its wording leans toward strong prices or activity; near -1 means it leans toward weakness. It is a count of phrases, not a full reading of the bank's intentions.

Model noteLatest scored BOJ statement has 16 hawkish vs 0 dovish phrases.

The world economic-news tone factor is +0.05 standard deviations, close to its recent norm; its common pattern explains 63% of the movement in the 7 series. It is dated 2026-10-06, 11 days before the last stored GDELT day (2026-10-17), because recent days are missing.

Simple exampleComputers read economic news stories from around the world and score how cheerful or gloomy the words are. This number blends those scores into one world mood reading. Above zero means the news sounds brighter than its recent normal; below zero means gloomier.

Model noteThe world economic-news tone factor is +0.05 sd, close to its recent norm (z-scores are vs each country's own trailing 2 years; PC1 captures 63% of the variance of the 7 series); it is dated 2026-10-06, 11 days before the last cached GDELT day (2026-10-17), because recent GDELT days are missing.

Typical rangeBetween -0.216 and 0.967 in half of all readings since Feb 8, 2019 (middle 50%; the outer 10th to 90th percentile band is -0.974 to 1.37). The latest reading sits at the 33th percentile.

Current world news tone is at or above 33% of the days observed since 2019-02-01. So about two-thirds of past days had a more positive tone than today.

Simple exampleThis places today's world news mood on a ladder of all the days the model has recorded. A reading of 80 would mean today sounds at least as upbeat as 80% of past days; a low reading means most past days sounded more positive.

Model noteCurrent world news tone exceeds or equals 33% of the days observed since 2019-02-01.

Typical rangeThis reading is itself a percentile of its own history, so its middle 50% is 25 to 75 by construction. The latest reading is 33.

Relative to each country's own two-year norm, US news tone is running better than the rest of the world, with a gap of +1.12. That is at the 94th percentile of its history, an unusually wide difference.

Simple exampleEach country's news mood is first graded against its own usual tone. This number then subtracts the average grade of five other big countries from the US grade. Above zero means US economic news sounds relatively brighter than the others; below zero means relatively gloomier.

Model noteRelative to each country's own two-year norm, US news tone is running better than the rest of the world (divergence +1.12; 94th percentile of its history).

Typical rangeBetween -0.514 and 0.472 in half of all readings since Jan 12, 2018 (middle 50%; the outer 10th to 90th percentile band is -0.898 to 0.855). The latest reading sits at the 95th percentile.

China has the weakest news tone compared with its own two-year norm, at -1.31 standard deviations. Its raw 7-day tone of +1.27 is still positive on GDELT's scale, because Chinese outlets tend to be more positive in level, which is why only the z-score is compared.

Simple exampleNames the country whose economic news sounds gloomiest compared with its own usual tone. Some countries' news always sounds upbeat or downbeat, the way some people always talk cheerfully, so each one is judged against its own habit rather than against the others.

Model noteCN has the weakest news tone relative to its own two-year norm (-1.31 sd; 7-day tone +1.27 on GDELT's scale).

The global geopolitical-risk index, measured from 10 US, UK and Canadian newspapers, is +0.63 standard deviations above its 10-year average in September 2026. That means somewhat more coverage of conflict and tension than usual.

Simple exampleLike counting how often newspapers mention wars, terrorism and international tension. Above zero means such stories are showing up more than usual; below zero means less. It measures how much the news talks about these risks, not how likely any event is.

Model noteGlobal GPR (measured from 10 US, UK and Canadian newspapers) is +0.63 sd above its 10-year trailing mean in the latest published month (2026-09).

Typical rangeBetween -0.241 and 1.14 in half of all readings since Jan 2017 (middle 50%; the outer 10th to 90th percentile band is -1.00 to 1.94). The latest reading sits at the 53th percentile.

Over 92 months, the world news-tone factor and the US sentiment factor from the sentiment model have moved together with a correlation of +0.54. That is a moderate positive link: they often agree, but not always.

Simple exampleShows how closely two mood gauges move together, like two friends whose moods tend to rise and fall on the same days. Close to +1 means they almost always move together, near 0 means no clear link, and below 0 means they tend to move in opposite directions.

Model noteOver 92 months the world news-tone factor co-moves +0.54 with the one-sided US sentiment factor from the sentiment module.

This compares the market value of US nonfinancial companies' stock with the size of the economy (GDP). A higher reading means stock values are large relative to what the economy produces in a year, a gauge that Warren Buffett once called one of the best single measures of valuation.

Simple exampleImagine the town's whole economy makes a certain amount of money in a year, and you add up what everyone would pay to buy all the local shops. This compares those two numbers. When the shops' price tag is much bigger than the yearly money the town makes, shop prices look stretched.

Model noteNonfinancial corporate equities stood at 255.0% of GDP in Q2 2026, the highest reading since the series begins in 1947 and above its 2000 peak of 162.6% of GDP and above its 2021 peak of 218.7% of GDP.

Typical rangeBetween 105.9 and 155.7 in half of all readings since Mar 31, 2000 (middle 50%; the outer 10th to 90th percentile band is 92.92 to 204.3). The latest reading sits at the 100th percentile.

This divides the market value of company stock by the economy-wide profits companies earn after tax. A higher reading means investors are paying more for each dollar of profit; a lower one means less. Because the top and bottom cover slightly different groups of companies, it is a rough gauge.

Simple exampleSuppose a lemonade stand earns $1 a year and you pay $10 to own it. You are paying 10 times its yearly earnings. This is that idea for all companies together. A bigger number means people pay more for each dollar of profit.

Model noteEquity value relative to economy-wide after-tax corporate profits was 21.4x in Q2 2026, at the 98th percentile of its history since 1947 and below its 2000 peak of 27.0x and above its 2021 peak of 20.2x.

Typical rangeBetween 12.13 and 16.82 in half of all readings since Mar 31, 2000 (middle 50%; the outer 10th to 90th percentile band is 10.11 to 19.91). The latest reading sits at the 96th percentile.

This is the portion of households' and nonprofits' financial assets that is held in corporate stock. A higher reading means more family wealth rides on stock prices; a lower one means more is held in cash, bonds and other assets.

Simple exampleThink of a family's piggy bank split into jars: cash, bonds and stocks. This shows how much of it sits in the stock jar. A fuller stock jar means the family's savings go up and down more when stock prices move.

Model noteCorporate equities made up 48.2% of household and nonprofit financial assets in Q2 2026, the highest reading since the series begins in 1945 and above its 2000 peak of 38.7% and above its 2021 peak of 41.0%.

Typical rangeBetween 28.55 and 35.14 in half of all readings since Mar 31, 2000 (middle 50%; the outer 10th to 90th percentile band is 25.70 to 40.70). The latest reading sits at the 100th percentile.

Margin debt is money investors borrow from brokers to buy securities. This measures how fast it grew over the past 12 months after removing inflation. Fast growth means more borrowing against stock holdings, which can make price swings larger in either direction; a negative reading means margin debt is shrinking.

Simple exampleMargin debt is like borrowing your big sister's money to buy more trading cards. If the amount borrowed grows fast, people are using more borrowed money. That can make prices swing bigger, up or down, because loans must be paid back.

Model noteInflation-adjusted FINRA margin debt changed by 26.7% over the 12 months to September 2026, at the 80th percentile of its history since 1998 and below its 2000 peak of 73.9% and below its 2021 peak of 67.2%.

Typical rangeBetween -9.21 and 21.76 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is -23.82 to 33.25). The latest reading sits at the 82th percentile.

The Shiller CAPE, from an academic source and not an official statistic, divides the stock market's price by its average inflation-adjusted earnings over the past ten years. A higher reading means investors are paying more for each dollar of long-run profits than they did in earlier periods; a lower one means less.

Simple exampleImagine judging a farm not by one year's harvest but by the average of ten harvests, so one lucky or unlucky year does not fool you. This ratio does that for the stock market. A bigger number means the price is high compared with long-run profits. It comes from a university professor's data.

Model noteThe Shiller CAPE (academic source, not an official statistic) was 40.7x in October 2026, at the 99th percentile of its history since 1881 and below its 2000 peak of 44.2x and above its 2021 peak of 38.6x (latest month provisional: Shiller uses a month-to-date price).

Typical rangeBetween 24.01 and 31.38 in half of all readings since Jan 2000 (middle 50%; the outer 10th to 90th percentile band is 21.00 to 37.34). The latest reading sits at the 97th percentile.

This is how much the Nasdaq-100, an index of large technology-heavy companies, has risen or fallen in price over the past two years. Researchers have studied cases where such an index more than doubled in two years, so the page checks how close the reading is to that run-up threshold.

Simple exampleThis is like checking how much a plant grew in two years. Researchers noticed that when a group of company shares more than doubled in two years, it was a special kind of fast run. This measures how far along that road the index is.

Model noteThe Nasdaq-100 two-year price change was 48.5% on 15 Oct 2026, at the 67th percentile of its history since 1988 and below its 2000 peak of 306.5% and below its 2021 peak of 115.5%; it is below the 100% two-year run-up threshold studied by Greenwood, Shleifer and You (2019).

Typical rangeBetween 7.66 and 51.83 in half of all readings since Jan 7, 2000 (middle 50%; the outer 10th to 90th percentile band is -25.07 to 78.46). The latest reading sits at the 72th percentile.

This is the share of the US economy spent on information-processing equipment and software, the category that includes computers, data centre hardware and software. A higher reading means businesses are putting a bigger slice of national output into this kind of investment.

Simple exampleIf a country is like a big household, this shows how much of its yearly spending goes on computers and software, the tools for building smart machines. A bigger share means more of the family budget is going to new tools.

Model noteInvestment in information-processing equipment and software was 5.1% of GDP in Q2 2026, the highest reading since the series begins in 1947 and above its 2000 peak of 4.5% of GDP and above its 2021 peak of 4.3% of GDP.

Typical rangeBetween 3.56 and 4.04 in half of all readings since Mar 31, 2000 (middle 50%; the outer 10th to 90th percentile band is 3.45 to 4.30). The latest reading sits at the 100th percentile.

This counts how many of the page's indicators are in the top tenth of their own history. A higher count means more of the measures look stretched at the same time; a lower count means they are mixed or ordinary. It describes conditions and is not a signal about timing.

Simple examplePicture a row of warning lights on a car dashboard, each lighting up when one gauge is in its highest tenth ever. More lights mean more gauges are high at the same time. The lights describe how things look; they do not say what comes next.

Model noteAs of 2026-10-18, 9 of 12 indicators are in the top decile of their own history (ai_capex_gdp, cape, concentration, excess_cape_yield, hh_equity_share, margin_gdp, mv_gdp, pe_econ, q_ratio); the equal-weight average percentile is 0.94.

This adds up what Microsoft, Alphabet, Amazon, Meta and Oracle spent on property and equipment such as data centres and servers over the latest four quarters. Higher means more is being invested in physical capacity. It counts all capital spending, not only AI.

Simple exampleImagine five big companies building new workshops and buying machines. This adds up everything they spent on workshops and machines over the last year. A bigger number means they are building more.

Model noteMicrosoft, Alphabet, Amazon, Meta and Oracle together spent $566 bn on property and equipment in the four quarters to 2026-06-30.

Typical rangeBetween 50.97 and 157.7 in half of all readings since Dec 31, 2014 (middle 50%; the outer 10th to 90th percentile band is 28.32 to 295.9). The latest reading sits at the 100th percentile.

This compares the latest four quarters of capital spending with the four quarters before. A positive reading means spending is growing; a negative one means it is shrinking. Large positive readings mean the build-out is expanding quickly.

Simple exampleIf you spent $10 on new toys last year and $20 this year, your toy spending doubled. This compares the last year with the year before for the five companies. Bigger means they are building faster than before.

Model noteThat trailing-year capital spending is 81% higher than in the four quarters a year earlier.

Typical rangeBetween 17.90 and 50.55 in half of all readings since Dec 31, 2015 (middle 50%; the outer 10th to 90th percentile band is 5.66 to 69.98). The latest reading sits at the 99th percentile.

This divides capital spending by the cash the five companies generate from running their businesses. Below 100 percent, operations cover the spending; above 100 percent, spending is bigger than operating cash, so other sources such as savings or borrowing must fill the gap.

Simple exampleSay you earn $100 from a paper route and spend $80 on a new bike. You used 80 percent of your earnings. If you spent $120, you would need savings or a loan for the other $20. This is that comparison for the five companies.

Model noteCapital spending used 82% of the five companies' operating cash flow; a reading above 100% would mean capex exceeded the cash their operations generated.

Typical rangeBetween 35.63 and 48.13 in half of all readings since Dec 31, 2014 (middle 50%; the outer 10th to 90th percentile band is 27.49 to 58.41). The latest reading sits at the 98th percentile.

This starts with operating cash, subtracts capital spending, then subtracts buybacks and dividends paid to shareholders. A positive amount means cash remained after all of that; a negative amount means spending and payouts went beyond what operations produced.

Simple exampleFrom your paper route money, you pay for a new bike and also give some to your friends as gifts. What is left in your pocket is this number. If it is below zero, you paid out more than you earned.

Model noteAfter capex, buybacks and dividends the five companies' combined cash flow was a surplus of $35 bn over the last four quarters.

Typical rangeBetween 26.46 and 55.01 in half of all readings since Sep 30, 2015 (middle 50%; the outer 10th to 90th percentile band is -10.50 to 81.24). The latest reading sits at the 39th percentile.

This is money the five companies raised by borrowing minus money they repaid, over the latest four quarters. A positive reading means debt grew from cash flows; a negative reading means they paid down more than they borrowed. Lease payments are left out.

Simple exampleThis counts money borrowed minus money paid back in a year. If you borrow $50 from a bank and pay back $20, you net borrowed $30. A bigger number means the companies took on more borrowing than they repaid.

Model noteDebt proceeds minus debt repayments in the five cash-flow statements, excluding finance-lease payments (including the $2.5 bn Alphabet reports inside its debt repayments), were $242 bn over the four quarters to 2026-06-30.

Typical rangeBetween -3.16 and 33.31 in half of all readings since Sep 30, 2015 (middle 50%; the outer 10th to 90th percentile band is -11.28 to 48.48). The latest reading sits at the 100th percentile.

This measures how much the group's long-term borrowing, such as bonds and term loans, has changed compared with a year earlier. Higher means debt is growing faster. Commercial paper and leases are not included, so it covers only part of total obligations.

Simple exampleIf you owed your parents $10 last year and owe $20 now, your debt doubled. This compares what the five companies owe in long-term loans now with a year ago. Bigger means the pile of loans is growing faster.

Model noteCombined long-term debt (notes and other borrowings including current maturities; commercial paper and leases excluded) was $484 bn, 100% more than a year earlier.

This compares Nvidia's sales over the latest four quarters with the four quarters before. Higher means its sales are growing faster. Nvidia is the leading supplier of AI chips, so its sales give a read on how much customers are buying.

Simple exampleThink of a lemonade stand that sold $10 of lemonade last year and $15 this year. This tells how much faster sales grew. Nvidia makes the chips that many AI computers use, so its sales hint at how much the buyers are buying.

Model noteNvidia's trailing-year revenue of $303 bn to its fiscal quarter ending 2026-07-26 is 83% above the prior year.

Typical rangeBetween 7.06 and 57.07 in half of all readings since Mar 31, 2011 (middle 50%; the outer 10th to 90th percentile band is -0.814 to 83.38). The latest reading sits at the 90th percentile.

Days sales outstanding shows how many days of sales are still waiting to be paid by customers. A higher number means customers take longer to pay. A rise alongside rising sales is something to watch, not proof of anything.

Simple exampleIf you sell cookies and your neighbours say they will pay next week, the number of days you wait for money is like this number. A long wait means customers take longer to pay. A long wait together with more sales is worth watching, but it is not proof of a problem.

Model noteNvidia's receivables equal 60 days of its latest quarter's sales; a rise alongside rising sales is a condition to watch for customer financing strain, not a conclusion.

Typical rangeBetween 37.59 and 50.54 in half of all readings since Mar 31, 2010 (middle 50%; the outer 10th to 90th percentile band is 33.89 to 56.98). The latest reading sits at the 93th percentile.

Compares the cash the companies' operations generated, after subtracting the value of shares handed out as employee pay, with the net income they reported, adding dollars across the group. A figure near or above 100% means reported profit was matched by cash; a lower figure means less of the profit showed up as cash over the same four quarters.

Simple exampleImagine a lemonade stand that says it earned $100 on paper, but only $90 of it is in the cash jar after you subtract the lemonade you gave your helper as payment. This number checks how much of the profit is really in the jar.

Model noteAcross 24 companies, operating cash flow minus stock-based compensation equalled 98% of reported net income over the four quarters to 2026-06-30.

Typical rangeBetween 124.2 and 155.4 in half of all readings since Mar 31, 2010 (middle 50%; the outer 10th to 90th percentile band is 119.0 to 166.4). The latest reading sits at the 0th percentile.

Adds each company's net new borrowing plus net share issuance (counted only where it is positive) and divides by the group's total capital spending. Higher means more of the building was matched by outside money rather than the companies' own cash, and the figure can exceed 100%.

Simple exampleSay you buy a $100 bike. You pay $40 from savings and $60 comes from a loan and from friends chipping in. This number asks what share of the building bill was paid by loans and new shares rather than money already earned.

Model noteNet borrowing plus net share issuance (counted only where positive) equalled 52% of the 26 companies' capital spending over the same period.

Typical rangeBetween 15.00 and 43.54 in half of all readings since Mar 31, 2010 (middle 50%; the outer 10th to 90th percentile band is 10.22 to 83.51). The latest reading sits at the 79th percentile.

Gains and losses on equity investments that companies record in income, as a share of their pre-tax income, for the companies that report them. These come from changes in the recorded value of stakes, not from selling products, so higher means more pre-tax income from that source and negative means net losses.

Simple exampleIf a card you own gets more valuable, you feel richer even though you have not sold it. Companies record gains like that on shares they own in other firms. This number shows how much of their income comes from that kind of price change.

Model noteGains and losses on equity investments recognised in income were 19.0% of pre-tax income for the 12 companies that tag them.

Typical rangeBetween 1.25 and 5.05 in half of all readings since Jun 30, 2010 (middle 50%; the outer 10th to 90th percentile band is 0.527 to 7.91). The latest reading sits at the 100th percentile.

Growth in customers' unpaid bills (receivables) minus growth in revenue over the past year, weighted by company size. A gap above zero means unpaid bills grew faster than sales, which can reflect longer payment terms or timing; near zero means the two moved together.

Simple exampleSuppose a shop's sales grow, but the amount of money friends still owe it for things sold on credit grows even faster. This gap compares those two growth rates. A big gap means more of the sales are still waiting to be paid.

Model noteRevenue-weighted, receivables grew 31.7 percentage points faster than trailing-year revenue across 25 companies.

Typical rangeBetween -0.029 and 6.53 in half of all readings since Jun 30, 2010 (middle 50%; the outer 10th to 90th percentile band is -2.90 to 9.93). The latest reading sits at the 100th percentile.

Total spending on property, equipment and data centres by the group over the latest four quarters, along with how it compares with a year earlier for companies reporting in both periods. Higher means a bigger build-out.

Simple exampleThink of everything a town spends in a year on new buildings and machines. This is the same kind of total for these companies, covering what they spent on chips, servers and data centres over the last year.

Model noteThe 26 companies reported $656 bn of capital spending over the four quarters to 2026-06-30, 74% more than a year earlier for the 26 companies reporting in both periods.

Debt the group raised minus debt it repaid, taken from the cash-flow statements, over the latest four quarters. Positive means total borrowing grew; negative means more was repaid than borrowed.

Simple exampleIt is the money you borrowed in a year minus the money you paid back. If you borrowed $50 and repaid $20, you now owe $30 more. This adds up the same thing for all these companies.

Model noteDebt proceeds minus repayments in the companies' cash-flow statements were $265 bn over the four quarters to 2026-06-30.

How much US commercial-bank lending to nondepository financial institutions (lenders and investment firms that are not banks) has grown versus a year earlier, after removing breaks that the Federal Reserve documents when banks reclassified loans. Higher means faster growth of this lending, which funds many kinds of nonbank lenders, not only data centres.

Simple exampleBanks lend money to other money-lending businesses, a bit like a big shop lending to smaller shops. This shows how much faster or slower that lending is than a year ago. It does not show what the money was used for.

Model noteCommercial-bank loans to nondepository financial institutions were $2064 bn, 19% above a year earlier after removing breaks documented in the H.8 notes (20% as reported; September 2026 average, released 2026-10-09).

Typical rangeBetween 10.06 and 18.29 in half of all readings since Jan 2016 (middle 50%; the outer 10th to 90th percentile band is 7.66 to 21.02). The latest reading sits at the 78th percentile.

How much private spending on building data centres in the US, from the Census Bureau, has changed versus a year earlier, measured as a seasonally adjusted annual rate. Higher means faster growth in construction, and the latest months are preliminary and get revised.

Simple exampleData centres are giant buildings full of computers. This compares how much was spent building them this year with the same time last year, like asking whether your town built more playgrounds than before.

Model notePrivate data-centre construction ran at a $85 bn annual rate, 73% above a year earlier (Census, August 2026, released 2026-10-01; preliminary).

Typical rangeBetween 17.91 and 56.09 in half of all readings since Jan 2015 (middle 50%; the outer 10th to 90th percentile band is 4.21 to 73.52). The latest reading sits at the 89th percentile.