JPM daily returns vs 10-year Treasury yield changes (last ~3 years)
The simple textbook claim — that banks “win” on days the 10‑year yield rises — doesn’t hold up cleanly for JPM when you look at daily moves over the past ~3 years. I compared JPM’s close-to-close returns on days the 10‑year rose versus fell and found average returns of +0.1568% on yield-up days versus +0.0644% on yield-down days; that raw gap is small and not statistically robust.
A linear fit and correlation tests show a positive but vanishingly weak relationship that explains almost none of daily return variance; one metric nudges toward significance while others are essentially null. Below you’ll find the full methodology, charts, and statistics that unpack why this is at best a negligible, inconsistent rate signal for JPM on a day-to-day basis.
For JPM over the past ~3 years, do trading days when the 10-year Treasury yield rises actually deliver stronger returns than days the yield falls — the textbook 'banks win when rates climb' belief? Thesis: JPM's daily returns show essentially no reliable link to the direction or magnitude of daily 10-year yield moves, debunking the idea that the stock is a clean bet on higher rates.
How this was measured
Resampled JPM minute bars to daily close-to-close returns. Reindexed the Treasury 10-year yield onto JPM's trading-day calendar with forward-fill and computed daily yield changes in basis points. Compared JPM returns on yield-up vs yield-down days (Welch two-sample t-test) and quantified linear association via OLS slope and Pearson/Spearman correlations. Flat (0 bp) days were excluded from the up/down buckets but counted for overall coverage.
The key numbers
Reading the numbers
Across 724 trading days (339 up, 341 down), JPM averaged 0.001568 on yield-up days vs 0.000644 on yield-down days — a tiny mean gap of 0.0009236 that is not statistically clear (two-sided p=0.4159).
The charts
The box plot shows large overlap between the return distributions on yield-up (n=339) and yield-down (n=341) days: means are 0.0016 versus 0.0006 and the central clouds clearly intersect. Note the extreme positive outlier on yield-up days (max 0.1047) and the deeper negative tail on yield-down days (min -0.0732), but those are isolated. The small difference in the group means is not backed up by a clear statistical separation (see p=0.4159), so the plot supports the idea that direction alone doesn't deliver a reliable payoff.
The scatter covers Δ10y from -19 to +19 bp with JPM returns spread vertically and no tight line: Pearson r ≈ 0.093 indicates a very weak linear link, even though the p-value is 0.012. The OLS slope is 0.0024125 per +10 bp (roughly a +0.24 percentage-point change in return for a 10 bp rise), but R² is only 0.0087, meaning the yield move explains under 1% of return variability. A rank check (Spearman ρ ≈ 0.005, p=0.889) further shows no robust monotonic relationship once outliers and ordering are considered.
This bar chart simply highlights the tiny mean gap: mean return 0.0016 on yield-up days versus 0.0006 on yield-down days. The raw difference (0.0009236) is visible here but small in absolute terms and, as the tests show, not statistically clear (two-sided p=0.4159). So the visual mean advantage on up days is real in number but not reliable enough to support a clean ‘‘banks win when rates climb’’ claim.
Directional return summary (JPM on yield-up vs yield-down days)
| direction | N | mean | median | std |
|---|---|---|---|---|
| Yield up | 339 | 0.0016 | 0.0017 | 0.0145 |
| Yield down | 341 | 0.0006 | 0.0018 | 0.0151 |
The takeaway
No — over the past ~3 years JPM does not show a reliable day-to-day win when the 10-year yield rises. On average JPM returned +0.16% on yield-up days (N=339) versus +0.06% on yield-down days (N=341), a raw mean difference of +0.092% that is not statistically significant (two-sided p ≈ 0.416). A simple linear fit gives a positive slope (~+0.241% per +10 bp) and Pearson r = 0.093 (Pearson p ≈ 0.012), but that relationship explains under 1% of daily return variance (R² ≈ 0.0087) and the rank test is essentially null (Spearman ρ ≈ 0.005, p ≈ 0.889). Put plainly: one test (Pearson) flags a tiny correlation that large sample size can make “statistically significant,” while other tests and the mean-difference t-test do not support a meaningful effect. This is at best a very weak, practically negligible signal, not a clean or actionable rate-driven payoff for JPM on a daily basis. In short: don’t treat JPM as a simple long-on-rates trade based on daily 10y moves — other factors dominate.
The fine print
- The 10y series was forward-filled onto JPM trading days (published with T+1 timing) — alignment could blur true intraday co-movement.
- Flat (0 bp) yield-change days (44 of 724) were excluded from the up/down comparison but included in totals; that choice doesn’t change the qualitative result.
- This is a bivariate, daily close-to-close study — earnings, macro headlines, or intraday dynamics are not controlled for and can mask or mimic rate effects.
- The window is ~36 months; different historical regimes could show different sensitivities.