Measure how two series move together, free and without signing in: a stock or ETF by its ticker or name, or an economic series by its id or title. Choose how often to sample them and how far back to look, and read their covariance, correlation and beta, with the data behind them.
This product uses the FRED® API but is not endorsed or certified by the Federal Reserve Bank of St. Louis.
Why covariance matters in finance
- Risk is a matrix, not a list. A portfolio's variance is the sum of every holding's variance and every pair's covariance. With 20 holdings there are 20 variances and 190 covariances: most of the risk is in the pairs.
- Diversification only works between assets that do not move together. Two assets with correlation 0.9 add little protection; a correlation near zero or below adds a lot.
- Hedges are covariances. A hedge ratio is a beta: the covariance of the position with the hedge divided by the hedge's variance.
- Relations change, and they change fastest when it matters. Correlations between risky assets tend to rise in sell-offs, just when diversification is needed. Some of that rise is mechanical (higher volatility alone raises a measured correlation); some is real.
- Macro data arrives late and is revised. A relation between a market and an economic release is only usable from the day the figure is published, and at the value first published.
How a lead works
- A lead is measured against the release, not the month described. Oil's change in January is set against the producer price index released in March, which describes February: a lag of 2 release months is a lead of 1 month of the period described.
- The usable time is to publication. From the end of the oil month to the release is 44 days at the median (331 releases since 1999): about 6 weeks.
From the Markets and Macro Quarterly, October 2026, Section 1.
How to use the calculator
The analyst: is the story in the data?
- Test a macro thesis before writing it. "Rising real yields hurt technology stocks": pick DFII10 and a technology stock or sector fund, weekly or monthly, 5 to 10 years. The correlation says whether the relation exists; the beta says how large it is; the window says whether it is recent or long-standing.
- Find what a company really moves with. A stock against oil, the dollar, the 10-year yield and credit spreads in turn: the strongest is its true macro driver, which is not always its sector label.
- Check that a relation is not two trends. Run it on changes (the default), never on levels: two rising series always look related as levels.
- Cite it. Every result names its data source; a pair, frequency and window can be shared as a link and reproduced by anyone.
Try an example: Exxon Mobil against WTI crude oil, daily, over 1 year
The portfolio manager: what does the book really hold?
- Diversification check before adding a position. The correlation of the candidate with the portfolio's largest exposure (or with SPY): near 1 adds concentration, near 0 or below adds diversification.
- Hidden factor exposure. Each holding against the 10-year yield, the dollar, oil or credit spreads: several holdings with large betas to the same series are one bet, whatever their sectors.
- Stress view. The same pair over a calm and a volatile window (for example 2024 against 2020): correlations usually rise in sell-offs, so the diversification that shows in calm data is the most a book can count on, not the least.
- Benchmark fit. A fund or ETF held in the book against its benchmark (for example a sector fund against SPY): the beta is the market exposure it actually carries, the R squared how much of its movement the market explains.
Try an example: a technology sector fund against SPY, daily, over 1 year
The trader: what has changed, and how much to hedge
- Hedge ratio. The beta of the position on its hedge instrument, over a look-back that matches the holding period (1 to 3 months for a short-term trade); size the hedge as position value times beta.
- Relative-value pairs. Two related names (KO and PEP, XOM and CVX): the correlation says whether the pair is still a pair; the beta gives the ratio to trade them in.
- Regime check. The same pair over 1 month, 3 months and 1 year: a short-window reading far from the long one is the kind of aberration this paper tests for, and the first sign that a hedge or a pair has stopped working.
- Macro event read. A sector against the series behind the next release (payrolls, CPI, claims), monthly, to see how strongly the sector has moved with that figure.
Try an example: Coca-Cola against PepsiCo, daily, over 1 month3 months1 year
From the Markets and Macro Quarterly, October 2026, Section 2.
What the Markets and Macro Quarterly, October 2026, found
Markets and Macro Quarterly, October 2026 (data through 2026-10-31). Each finding links to its section of the issue; the calculator opens on the pair behind it.
US stocks no longer tell you what Tokyo will do next morning. When US energy stocks rose during the New York day, Japan's Nikkei 225 tended to rise at the next open; over the last 12 months that link fell from +0.42 to −0.06, the largest 12-month drop on record. Section 4.4
Try this pair: an energy sector fund against Japan's Nikkei 225, daily, over 1 year2 yearsThe calculator measures the same day; the next-morning link is the study's own test.
A “safe” part of the stock market has been behaving like a bond. Consumer Staples shares have been falling when interest rates rise, more sharply than in 99% of their history since 1999; part of this is a more volatile market rather than a lasting change. Section 4.2
Try this pair: a Consumer Staples sector fund against the 30-year Treasury yield, daily, over 1 month1 year
Treasuries stopped cushioning stock falls. Since the Fed began raising rates in 2022, the S&P 500 and the 10-year yield went from moving clearly together (+0.36) to about no relation (−0.10), so falling stocks no longer bring rising bond prices. Section 4.7
Try this pair: SPY against the 10-year Treasury yield, daily, over 1 month2 years
Bonds and foreign holdings now tend to lose at the same time. Rising US rates now come with a stronger dollar (−0.08 before, +0.31 since 2021), so when rates jump, bonds fall and the stronger dollar shrinks the value of unhedged foreign holdings. Section 4.7
Try this pair: the 10-year Treasury yield against the broad dollar index, weekly, over 5 years20 years
Real estate stocks now track mortgage rates far more closely. The link went from weak (−0.11 to 2020) to strong (−0.62 in 2021–23, −0.51 since). Section 4.8
Try this pair: a real estate sector fund against the 30-year mortgage rate, weekly, over 1 year2 years
What the study expected and did not find
Expected, not found: a turning point at the pandemic. The data date the stock-bond turn to 1 April 2022 and the rate-dollar turn to 30 March 2022, when the Fed began raising rates. Section 4.7
Expected, not found: economic data that warns ahead of US stock moves. None of the indicators tested leads US stocks once clustered calm and stormy spells are allowed for; across the study, 1,280 of 1,956 findings of the ordinary test no longer held. Section 4.9
Expected, not found: energy as an early signal. Of 23 energy leads tested, only one kind held: the price of oil gives about two months' notice of the next US consumer price figure. Section 4.3
Try this pair: WTI crude oil against US consumer prices, monthly, over 10 yearsThe calculator measures the same month; the two-month lead is the study's own test.
Expected, not found: relationships breaking in a turbulent year. Over the last 1, 3 and 6 months nothing moved outside its own history by more than chance alone would produce (139 flags against 119 by chance). Section 4.2
Explore your own ideas for free: every relation in this paper, and any pair of your own, can be measured in Gyre's free Covariance Calculator at gyreresearch.com/covariance.
Read the full issueDownload the PDFSubscribe free
Not investment advice, and not a recommendation to buy or sell anything.
See these measures at work: Markets and Macro, Gyre Research's quarterly, monthly and weekly publications on how markets and the economy move together. Read the latest issue and subscribe free.
The calculator
Levels over the window
Paired period values
Daily table: 17 pairs
Computed November 17, 2026. Data through November 16, 2026. Daily changes (two-day changes where marked), computed by this calculator after each close. Select a pair to open it in the calculator.
| Correlation over the last N changes | 1-year look-back | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Pair (A/B) | 1 month (21) | 3 months (63) | 1 year (252) | Beta of A on B | R squared | St. dev. A | St. dev. B | Window | Observations |
| ETFs | |||||||||
| SPY/TLT | -0.07 | 0.10 | 0.26 | 0.37 | 0.07 | 0.87% | 0.61% | 2025-11-17 to 2026-11-16 | 251 |
| SPY/GLD | 0.36 | 0.39 | 0.36 | 0.17 | 0.13 | 0.87% | 1.82% | 2025-11-17 to 2026-11-16 | 251 |
| SPY/QQQ | 0.96 | 0.94 | 0.93 | 0.59 | 0.87 | 0.87% | 1.36% | 2025-11-17 to 2026-11-16 | 251 |
| QQQ/IWM | 0.84 | 0.77 | 0.75 | 0.86 | 0.56 | 1.36% | 1.18% | 2025-11-17 to 2026-11-16 | 251 |
| SPY/EFA | 0.85 | 0.80 | 0.80 | 0.66 | 0.63 | 0.87% | 1.05% | 2025-11-17 to 2026-11-16 | 251 |
| SPY/EEM | 0.91 | 0.83 | 0.79 | 0.41 | 0.62 | 0.87% | 1.68% | 2025-11-17 to 2026-11-16 | 251 |
| SPY/HYG | 0.65 | 0.62 | 0.74 | 2.57 | 0.54 | 0.87% | 0.25% | 2025-11-17 to 2026-11-16 | 251 |
| TLT/GLD | 0.17 | 0.36 | 0.16 | 0.05 | 0.03 | 0.61% | 1.82% | 2025-11-17 to 2026-11-16 | 251 |
| ETF against a FRED series | |||||||||
| SPY/VIXCLS two-day | -0.98 low | -0.94 low | -0.85 | -0.09 | 0.73 | 1.19% | 11.01% | 2025-11-18 to 2026-11-13 | 249 |
| SPY/DGS10 two-day | 0.56 | 0.21 | -0.17 | -0.03 | 0.03 | 1.19% | 0.06 | 2025-11-18 to 2026-11-13 | 247 |
| SPY/DCOILWTICO two-day | -0.35 | -0.43 | -0.38 | -0.09 | 0.15 | 1.19% | 5.30% | 2025-11-18 to 2026-11-11 (behind) | 247 |
| GLD/DTWEXBGS two-day | -0.01 | -0.31 | -0.49 | -3.24 | 0.24 | 2.53% | 0.38% | 2025-11-18 to 2026-11-06 (behind) | 243 |
| Stocks | |||||||||
| AAPL/MSFT | 0.55 | 0.43 | 0.16 | 0.12 | 0.02 | 1.57% | 2.14% | 2025-11-17 to 2026-11-16 | 251 |
| NVDA/AMD | 0.57 | 0.49 | 0.51 | 0.27 | 0.26 | 2.45% | 4.64% | 2025-11-17 to 2026-11-16 | 251 |
| XOM/CVX | 0.95 | 0.92 | 0.87 | 0.95 | 0.75 | 1.71% | 1.55% | 2025-11-17 to 2026-11-16 | 251 |
| JPM/BAC | 0.94 | 0.72 | 0.77 | 0.78 | 0.59 | 1.43% | 1.41% | 2025-11-17 to 2026-11-16 | 251 |
| KO/PEP | 0.80 | 0.71 | 0.62 | 0.54 | 0.38 | 1.19% | 1.37% | 2025-11-17 to 2026-11-16 | 251 |
- SPY/VIXCLS: the 1 month (21-change) correlation -0.98 is below its 1st percentile, -0.97, of 718 earlier 21-change correlations, against its history since December 5, 2023.
- SPY/VIXCLS: the 3 months (63-change) correlation -0.94 is below its 1st percentile, -0.92, of 676 earlier 63-change correlations, against its history since December 5, 2023.
Correlations over the last 21, 63 and 252 changes (1 month, 3 months and 1 year of trading days), computed from the calculator's own aligned changes over all shared history. A 1-month or 3-month value marked high or low is above the 99th or below the 1st percentile of every earlier correlation of that pair over the same number of changes, against its history since December 4, 2023 (SPY/VIXCLS, SPY/DGS10, SPY/DCOILWTICO, GLD/DTWEXBGS: since December 5, 2023), about 3.0 years.
Beta, R squared, standard deviations, window and observations: the calculator's own result over a 1-year look-back. Standard deviation of each change: a percent for prices and series not measured in percent; in percentage points for a series measured in percent (its change in level).
two-day SPY/VIXCLS, SPY/DGS10, SPY/DCOILWTICO, GLD/DTWEXBGS: a security's 4 pm close against a FRED series recorded at another time of day, so the pair is measured on overlapping two-day changes (each spans two trading days, so a move one series records a day after the other still falls in a shared change). Its standard deviations are of two-day changes and its covariance is annualized by 126 (252 / 2).
- SPY/DCOILWTICO: its data ends November 11, 2026, 3 business days before the newest pair; the source had not published later values when this table was computed.
- GLD/DTWEXBGS: its data ends November 6, 2026, 6 business days before the newest pair; the source had not published later values when this table was computed.
- SPY/VIXCLS: VIXCLS is a FRED series: FRED's daily values are not all taken at the 4 pm New York stock close (the broad dollar index, DTWEXBGS, uses noon New York rates) and skip their own holidays, which pulls a daily correlation toward zero. Overlapping two-day changes are used here, so a move one series records a day after the other still falls inside a shared change.
- SPY/VIXCLS: Two-day changes overlap: each value is the change over the last two trading days, so consecutive values share a day. The covariance per period is two days' worth, so the annualized covariance is the per-period figure x 126 (252 / 2). Overlapping values carry about the information of half as many separate ones, so the estimates are less precise than 249 observations suggest.
- SPY/DGS10: DGS10 is a FRED series: FRED's daily values are not all taken at the 4 pm New York stock close (the broad dollar index, DTWEXBGS, uses noon New York rates) and skip their own holidays, which pulls a daily correlation toward zero. Overlapping two-day changes are used here, so a move one series records a day after the other still falls inside a shared change.
- SPY/DGS10: Two-day changes overlap: each value is the change over the last two trading days, so consecutive values share a day. The covariance per period is two days' worth, so the annualized covariance is the per-period figure x 126 (252 / 2). Overlapping values carry about the information of half as many separate ones, so the estimates are less precise than 247 observations suggest.
- SPY/DCOILWTICO: DCOILWTICO is a FRED series: FRED's daily values are not all taken at the 4 pm New York stock close (the broad dollar index, DTWEXBGS, uses noon New York rates) and skip their own holidays, which pulls a daily correlation toward zero. Overlapping two-day changes are used here, so a move one series records a day after the other still falls inside a shared change.
- SPY/DCOILWTICO: Two-day changes overlap: each value is the change over the last two trading days, so consecutive values share a day. The covariance per period is two days' worth, so the annualized covariance is the per-period figure x 126 (252 / 2). Overlapping values carry about the information of half as many separate ones, so the estimates are less precise than 247 observations suggest.
- GLD/DTWEXBGS: DTWEXBGS is a FRED series: FRED's daily values are not all taken at the 4 pm New York stock close (the broad dollar index, DTWEXBGS, uses noon New York rates) and skip their own holidays, which pulls a daily correlation toward zero. Overlapping two-day changes are used here, so a move one series records a day after the other still falls inside a shared change.
- GLD/DTWEXBGS: Two-day changes overlap: each value is the change over the last two trading days, so consecutive values share a day. The covariance per period is two days' worth, so the annualized covariance is the per-period figure x 126 (252 / 2). Overlapping values carry about the information of half as many separate ones, so the estimates are less precise than 243 observations suggest.
How well do the estimates forecast?
For each of 17 pairs, at every day with a year of earlier changes, three estimates of the covariance were made and compared with what the next 21 trading days then showed. A loss of 0 would be a perfect forecast; lower is better. The comparison with last year's sample is a Diebold-Mariano test (t of -2 or lower = clearly better).
| Estimate | Average loss | Error size against last year's | Pairs with a smaller error | Test t |
|---|---|---|---|---|
| Last year's sample (the table's estimate) | 0.970 | 1.000 | – | – |
| Exponentially weighted, decay 0.94 | 0.930 | 1.140 | 0 | -0.32 |
| Last 3 months' sample | 0.982 | 1.098 | 0 | 0.15 |
Exponentially weighted estimate offered: no: mean loss 0.9300 against 0.9702, DM t -0.32, lower covariance RMSE in 0 of 17 pairs
Latest in Markets and Macro
Markets and Macro Quarterly, October 2026
Released · data through
What this issue says about this tool:
- Section 2: How to use the calculator
- US stocks no longer tell you what Tokyo will do next morning.
- A “safe” part of the stock market has been behaving like a bond.
- Treasuries stopped cushioning stock falls.
- Bonds and foreign holdings now tend to lose at the same time.
- Real estate stocks now track mortgage rates far more closely.
- Expected, not found: energy as an early signal.
Method
Every number on this page comes from the steps below, in this order. Nothing is smoothed, filled in or estimated.
- Data. A security's price is its daily adjusted close (the close for currencies and crypto, which have no adjusted close). Our price history begins on 2023-12-01, so a security can be looked back at most to that date. A FRED series is downloaded from the Federal Reserve Bank of St. Louis the first time anyone uses it and refreshed when FRED reports an update; FRED's "no value" marks are left out.
- Periods. Each series is put into periods at the chosen frequency (the day; the week, Monday to Sunday; the calendar month; the quarter; the year) and the last value in each period is kept. A weekly period is shown as its Friday, the others as their last day.
- Alignment. Only periods in which both series have a value are kept, so both changes are always measured over the same interval. A period that is not over by the end of the window (for example the current month) is left out and named in the notes.
- Changes. Each series is measured between consecutive kept periods as a percent change, a change in level, a log change, or left as its level. Percent and log changes need values above zero. A two-day change (daily frequency only, for both series) is measured from each day to the day two kept days later, as a percent change or, for a FRED series measured in percent, a change in level; the changes overlap, one per day. It is meant for two series recorded at different times of day (a stock's 4 pm close and a rate or currency fixed earlier): a move one series records a day after the other still falls inside a shared change.
- Statistics. Sample covariance with the N−1 denominator, Pearson correlation, R² (the square of the correlation), and each beta (the covariance divided by the other series' variance). The annualized covariance multiplies the per-period figure by the periods in a year: 252 trading days, 52 weeks, 12 months, 4 quarters or 1 year. A two-day change holds two days' worth of covariance per period, so it is annualized by 126 (252 divided by 2). Levels are not annualized.
Rules the calculator enforces
- A series cannot be used more often than it is published: a monthly FRED series needs a monthly, quarterly or annual frequency.
- At least 10 shared observations are needed. With fewer than 30 the page says the estimate is imprecise and gives the correlation's approximate standard error.
- The look-back menu offers only the periods both picked series can fill; longer ones are greyed out, and "All shared history" is always available.
- "2 weeks" and "1 month" count trading days, not the calendar: the last 10 or 21 daily changes both series share, ending on the latest day they share. They need daily frequency. With 10 changes, 11 shared days are needed (10 is the minimum).
- The two-day change needs daily frequency and measures both series.
- When one series starts later than the look-back, the window starts where both have data, and the page says so.
Reading the result
Covariance is in the units of the two measures multiplied together (for two percent changes, a fraction squared), so its size depends on the series. Correlation removes the units and runs from −1 to 1. The beta of the second series on the first is how much the second series' change moves, on average, for a one-unit change in the first.
FRED data and its owners
Each FRED series is listed with its source under the result. FRED marks some series as owned by a third party that requires permission before its data is used; the calculator does not offer those, nor series FRED gives no copyright status.
Economic data: Federal Reserve Bank of St. Louis, FRED®.
SPY against the VIX carries both a low 21-change and low 63-change mark: the inverse link is the most negative in its roughly three-year history.
On November 17, 2026, the table's only marks are on SPY against VIXCLS, flagged low on both the 21-change and 63-change windows because the inverse relationship is more negative than at any comparable point since its history began in December 2023. Every other pair sits inside its own 1st-to-99th percentile range. Compared with November 14, 2026, one-year correlations and betas moved only slightly across all 17 pairs. Several pairs show short-window correlations that differ markedly from their one-year figures, including a near-zero dollar-gold reading and a positive recent SPY-yield reading against a negative annual one.
These figures measure co-movement only, from data beginning in December 2023, and show nothing about what drives any pair or where correlations go next.
This analysis was created with the assistance of large language model technology. It is written from this page's numbers for the day and describes the results; it is not a forecast or investment advice.
Past daily commentary