Loading…
FX BANK FORECAST · COVERAGE
Institutional FX coverage in your inbox
Aggregated year-end forecasts, scenario shifts, and curated analyst notes from 38 institutional desks. No promotion.
FX BANK FORECAST · COVERAGE
Aggregated year-end forecasts, scenario shifts, and curated analyst notes from 38 institutional desks. No promotion.
J.P. Morgan Global Research has launched a systematic model for trading 6Mx10Y EUR rates gamma, detailed by Khagendra Gupta and Marcus Mayfield in a podcast recorded 6 November 2026, per the desk's note . The product sits squarely at the intersection of EUR rates vol and the front-end/ belly slope, and the desk's implicit claim is that gamma in the 6M-into-10Y sector is structurally mispriced relative to realised vol and can be harvested rule-based rather than discretionarily. The supporting evidence is thin in the public-facing excerpt — JPM directs institutional clients to the underlying repo (GPS-5523754-0) for the actual signal construction, backtest statistics, and parameter set — so the desk-of-record read has to lean on the model's framing rather than a quoted number. There is no EUR/USD or cross-FX view embedded here; this is a rates-gamma product that spills over into EUR swaption positioning and dealer hedging flows rather than a directional currency call. With no tracked pair attached and no high-impact Eurozone or US event in the next 30 days within our calendar window, the actionable content is the framework itself, not a trade trigger.
Per the full note , J.P. Morgan's rate strategy team — Khagendra Gupta and Marcus Mayfield — has built a systematic model to trade 6Mx10Y EUR rates gamma, i.e. the 6-month-expiry option on the 10-year EUR swap rate. The framing is a rules-based approach to a sector of the EUR swaption surface that discretionary desks typically handle by feel, and the desk is presenting the model as a repeatable way to capture gamma rather than a one-off idea.
The public excerpt carries no backtest numbers, Sharpe, or vol-realisation stats — JPM gates the substance behind the institutional repo link (GPS-5523754-0), which is standard for their podcast-plus-report format. That absence is itself informative: the podcast is a teaser, and the model's actual edge (signal construction, entry/exit rules, sizing) is in the gated document. The alternative read — that 6Mx10Y gamma is simply too cheap or too rich on a realised-vs-implied basis and needs no systematic overlay — is what the desk is implicitly rejecting by building a model at all.
There is no cross-firm read to anchor against here: our coverage bundle holds no per-firm EUR rates or EUR/USD targets for this commentary, so no aligned or contrary grouping is possible. What matters instead is the adjacent complex — EUR/USD, the ECB deposit rate path, and Bund swap spreads — because 6Mx10Y swaption gamma is where dealer hedging of EUR rates vol transmits into spot and cross-market flows. Watch EUR/USD and the front-end ECB pricing as the transmission channels for any gamma-driven hedging the model would generate.
No high-impact Eurozone or US events sit on our 30-day calendar for this jurisdiction, so the model's signal generation is unencumbered by scheduled event risk in the immediate window. That makes the current period a relatively clean test bed for the systematic framework, but it also means there is no near-term catalyst to force the gamma trade on or off — the model's own rules, not the data calendar, will drive positioning.
Key takeaways
Market implications
Watch 6Mx10Y EUR swaption implied vol against realised 10Y swap-rate vol — the gap is the model's raw material — and monitor EUR/USD around 1.07 as the transmission level where gamma-hedging flows would show up in spot. Dealer positioning in EUR rates vol is the signal to track, since systematic gamma strategies change how the street hedges its swaption book.
Risks to this view
A vol shock that repriced 6Mx10Y EUR gamma far outside the model's historical parameter range would invalidate the systematic edge and force a rules-based exit. An ECB pivot or surprise ECB communication — even off-calendar — would reset the 10Y swap-rate realised-vol regime, breaking the realised-vs-implied relationship the model relies on. Finally, crowded positioning in EUR rates vol would compress the very gamma premium the strategy is designed to harvest.
Hi, and welcome to At Any Rate, J.P. Morgan's global research podcast, where we take a look at some of the drivers behind the biggest trends and themes across fixed income currencies and commodity markets. I'm Kajendra Gupta from the European rate strategy team at J.P.
Morgan, and today I'm joined by my colleague, Marcus Mayfield, to discuss systematic strategy in trading euro rates, the rate scam in six-month tenure. Specifically, we'll discuss our newly published model, which significantly improves performance compared to our previous model and is generally reliable in anticipating future gamma returns. As a background, systematic funds are a big driver of rates volatility across TM, and specifically, euro rates.
They are typically present across the surface, but focus generally on liquid gamma points. The basic idea is to monetize the typical premium that options markets have in the form of higher implieds compared to exposed delivered volatility. On average, an always short gamma strategy is profitable over long horizons.
This is a slow-earning strategy which exploits this implied delivered volatility differential. However, periods of uncertainty and associated high volatility could quickly erode the returns from an always short vol strategy. Thus, a strategy that can switch from short to long during these volatile periods is highly desirable in my view.
So with this background, Marcus, let's get down to your new strategy. Can you tell us what this strategy is and how the results are looking? Thank you for the introduction, Kajendra.
Our new strategy applies a sign-constrained bridge-regularized logistic regression to a range of indicators that we believe drive six-month, 10-year euro gamma returns. That's quite the mouthful, so let's break it down piece by piece. The logistic regression converts our indicators into a probability that being long rather than short gamma will first meet the profit target that we have set for a trade.
Sign constraints then ensure the coefficients estimated by our model are economically sensible. Take a rapid deterioration in confidence survey scores. All else equal, this should increase uncertainty around the central bank's policy rate path and so favor gamma longs over shorts.
The sign constraint makes the model respect this. A sharp deterioration in confidence can raise the probability that long gamma is the right position or leave it unchanged but never lower it. Finally, the bridge regularization guards against overfitting by penalizing large coefficients.
This three-pronged approach distills various drivers of euro gamma returns into a single number while simultaneously supporting model stability and keeping the model intuitive in conditions it has not seen before. Now, shifting focus to the results, our backtest shows the new strategy performing well overall and through different regimes. Since 2004, the strategy has delivered a net information ratio of 1.1, an average annualized return of around 110 basis points of notional and a 57% hit rate.
It has also performed well in periods that were favorable for gamma shorts like 2009 to 2013 and periods where gamma longs benefited such as 2008 and 2022. So far, performance has also held up in 2026 with successful longs and shorts contributing to an annualized net return around 120 basis points of notional. Okay, that is definitely a mouthful in terms of the nomenclature and the strategy does look attractive in terms of results.
But if I were to dig a bit further, the success of any such strategy in my mind depends on the choice of ex-ante indicators. So what type of indicators have you used to devise this strategy? Great question.
Our strategy uses 16 indicators which we group into three categories. One, macro regime, two, euro rates for state and transmission, three, cross-asset sources of risk. The first group captures the macro variables that define the structural regime, inflation, growth, and then momentum.
The second measures how far uncertainty transmits to the euro rates complex. It includes volatility across the money market curve, momentum in intra-EME spreads, and as a measure of valuation, the level of six-month, 10-year euro invites. The third covers cross-asset sources of risk, both domestic and external.
Domestic sources include euro area, equity, credit, and FX volatility, and external sources include six-month, 10-year dollar invites and the VIX. Okay. So as a starting point, these seem like a lot of indicators to deal with for me.
Are all of these statistically significant? And more importantly, how do you consolidate all these information coming from these various indicators into one trading signal? See, that's quite right.
The model does use 16 indicators in total. Rather than test each in isolation, we assess their contribution by adding indicator groups incrementally to a backtest. The macro variables, which define the structural regime, form the baseline.
They perform well during major macro episodes, such as the global financial crisis, the recovery following the euro zone debt crisis, and the post-COVID inflation spike, but are muted otherwise. Adding the euro rates fault group improves on this by capturing events like the May 2015 fund far shock, as well as recent volatility. Then adding cross-asset risk indicators improves the results further.
The strategy is positioned correctly for global shocks, such as the risk of moves of March 2020, and additionally avoids frequent smaller drawdowns. As discussed earlier, a logistic regression converts these indicators into a single probability that a long gamma position hits our profit target. Because many indicators are highly correlated, we only use some of them in levels.
For most, we instead calculate a velocity score that measures how rapidly the indicator is moving relative to its recent history. This isolates the information we want each indicator to carry and better captures short-term inflection points where market variables can act as early warning or contemporaneous stress signals for euro gamma returns. Grid regularization further spreads the model weights across correlated indicators so that no single indicator dominates.
Finally, we map the probability to a score between minus one, meaning a maximum conviction short, and plus one, a maximum conviction long, which is our trading signal. We then optimize the model and trading parameters through backtesting, and the resulting strategy goes long gamma when a signal rises above a positive 0.175 conviction threshold and stays long gamma until the signal turns negative. It goes short when the signal falls below the corresponding negative 0.175 threshold, stays short until the signal then turns positive.
Okay, that makes sense, especially transforming the variable to ensure limited cross-correlation. I was particularly impressed by the observation that the signal switches quickly when macro conditions change, and thus has led to a reasonable outperformance versus an always short gamma. I know we have gone through a lot of details in both in terms of performance and technicals, but here are my final two questions.
What is the strategy currently pointing to? Long, short, neutral, and then now regular readers of our research pieces may remember that we had published another systematic trading rule a couple of years ago. So how is this strategy different from that one, both in terms of methodology and performance?
So the strategy is currently short, six-month, 10-year Euro gamma, having switched between long and short several times this year. For most of the year, this signal has reflected a tussle between the severity of potential shocks as measured by our Euro rates and cross-asset indicators and the level of six-month, 10-year implieds. When large rapid increases in implied and delivered volatilities have outweighed the indicators pointing short, the signal has risen above the conviction threshold, and the strategy has gone long.
As each shock has subsided and implieds have become rich, the signal has turned negative and the strategy has flipped short. September was a clear example. A sharp rise in the velocity of 10-year swap yields and our measure of cross-sectional money market volatility pushed the strategy to enter a long gamma position on the 10th of September.
Implieds then ground higher over the following weeks as uncertainty over the Middle East and intra-EMU spreads persisted. But as the initial shock faded and implied stayed elevated, the short contribution from rich implieds came to outweigh the long contributions elsewhere, and the strategy flipped short on the 8th of October. That dynamic is still at play, but the short contribution from implieds has shrunk as they have fallen over the past month, even though the signal remains negative and the strategy short.
To show how this might change, we outline two purely hypothetical scenarios. These only illustrate how the signal works and do not represent market views. First, if implieds were to keep falling and broader market volatility were then to pick up, this would, all else equal, push the signal towards long gamma.
Second, if implieds were to cheapen enough, their contribution would shift in the long direction, even with market volatility muted. If the signal did turn positive, the strategy would move to neutral and it would only turn long if the signal rose above positive 0.175 threshold. Turning to your second question, our previous approach used a random forest classifier trained on eight macroeconomic medical indicators.
It built 1,000 decision trees, each casting a vote of plus one, so long gamma, or minus one, short gamma, and then took the average of those votes as a signal. Like our new signal, it therefore ranged from minus one to plus one, and we apply conviction thresholds to it to form a trading strategy. The key difference is how the signal is built.
Rather than averaging discrete votes, the new model estimates the probability that a long gamma position hits our profit target and maps it to a score. The new model also draws on a broader set of 16 indicators, and unlike the old model, which used all of its indicators and levels, transforms most of them into these velocity scores. Additionally, the new model has sign constraints on each variable, whereas the previous model was free to assign non-intuitive signs as long as the profit target was met.
Overall, as we show in our analysis, the new strategy convincingly outperforms the random forest strategy, both on a risk-adjusted basis and in terms of cumulative P&L. We also find statistical evidence that the new strategy signal reliably anticipates future gamma returns, with stronger signals associated with larger average gains. Thanks, Marcus, for that.
Well, that's all from us. Thank you for listening, and stay tuned for more updates on the fixed income space here on At Any Rates, J.P. Morgan's global research podcast series.
This communication is provided for information purposes only. Please read the J.P. Morgan research reports related to its contents for more information, including important disclosures.
Copyright 2026, J.P. Morgan Chase & Company, all rights reserved. This episode was recorded on 6th November, 2026.
How we cover this story
Live cross-firm bank consensus across 38 desks — FX, oil & gold
View bank forecasts