07 — Academic FX Strategies Beyond the Baseline Three

Context. TSMOM, Asian-range mean reversion, and pre-fee carry have already been tested on 15 years of real data and produced no edge. This document surveys the next tier of academic strategies — the ones that have survived peer review and repeated out-of-sample tests by independent authors — so we can decide which deserve implementation.

Stance. Skeptical by default. A Sharpe of 1.0+ in a JFE paper means almost nothing unless (a) it survives transaction costs, (b) survives the post-publication decade, and (c) doesn't require data that retail can't actually buy. Every strategy below is graded on those three filters.

Currency excess return convention throughout: long high-yielder funded by short low-yielder, or long undervalued vs short overvalued, etc. — i.e. zero-cost carry-style portfolios. All Sharpes are annualized.

Strategies that have evidence behind them

1. Cross-Sectional Currency Momentum (CS-MOM)

• Paper. Menkhoff, Sarno, Schmeling, Schrimpf (2012), Currency Momentum Strategies, Journal of Financial Economics 106(3), 660–684. Working paper: BIS WP 366 / SSRN 1988679.
• Mechanic. Each month, rank ~48 currencies (vs USD) by total return over the past 1–12 months (formation period f). Go long the top quintile, short the bottom quintile, hold for 1–12 months (holding period h). The strongest combinations are short-formation/short-holding (1/1, 3/1, 6/1). This is cross-sectional, not time-series — you always hold the winners vs losers, irrespective of the absolute trend.
• Reported. Spread up to ~10% p.a. between top and bottom portfolio. Implied gross Sharpe in the paper is roughly 0.9–1.0 for the best (1/1) sort, ~0.6–0.7 for the more tradable (3/1, 6/1) sorts. Max drawdown not headlined in the paper; in independent reconstructions it has reached -25% to -35% in 2008–2009 and 2014–2015.
• Data. Yes, retail-accessible: spot + forward rates for ~30–40 liquid currencies. Bloomberg or Refinitiv ideal; Dukascopy/HistData adequate for G10+majors.
• Honest net Sharpe. 0.2–0.4 for a retail trader running G10+EM with realistic bid/ask. The paper itself flags that "transaction costs reduce returns substantially" and that the strategy concentrates in less liquid EM currencies where the spread is largest — exactly where retail bid/asks are punitive. Verdad (systematic FX) reports the public momentum benchmark went negative post-2010.
• Failure mode. Crowded factor. Most published edge sits in EM currencies that you can't get tight spreads on. Performance has decayed badly in the QE / suppressed-vol era (2012–2020).

2. Dollar Carry / Dollar Factor (Time-Series Carry on the USD)

• Paper. Lustig, Roussanov, Verdelhan (2014), Countercyclical Currency Risk Premia, JFE 111(3), 527–553. NBER WP 16427 | SSRN 1541230.
• Mechanic. Each month, compute the average forward discount of a basket of foreign currencies vs USD. If the average foreign currency has a higher yield than USD (average forward discount > 0), go long the equally-weighted basket of foreign currencies and short the USD. Reverse when USD is the higher-yielder. This is a time-series, dollar-side carry — orthogonal to the standard cross-sectional carry that's been arbitraged down.
• Reported. Sharpe ~0.6–0.8 in-sample (1983–2010). Annual return ~5–7% at typical 8–10% vol. The forward discount + U.S. industrial production growth jointly forecast up to 25% of one-year dollar return variation.
• Data. Yes. You need (a) USD forward rates vs G10, (b) U.S. industrial production (FRED, free). All accessible.
• Honest net Sharpe. 0.3–0.5. This one is genuinely different from the carry the user already tested. Crucially it has a macro conditioning variable (US IP growth) that ties it to the business cycle rather than purely to a static yield rank. Out-of-sample (post-2014) performance has been mediocre but not catastrophic — better than cross-sectional carry over the same window.
• Failure mode. When the Fed is in an unusual regime (ZIRP 2009–2015, the 2022–2023 hiking cycle that broke historical correlations), the dollar-side signal whipsaws.

3. FX Value / PPP Reversion

• Paper. The canonical implementation paper is Kroencke, Schindler, Schrimpf (2014), International Diversification Benefits with Foreign Exchange Investment Styles, Review of Finance 18(5), 1847–1883. SSRN 1790764. Theoretical foundations in Asness, Moskowitz, Pedersen (2013), Value and Momentum Everywhere, Journal of Finance 68(3), 929–985. SSRN 1363476 | AQR paper.
• Mechanic. For each currency, compute deviation from PPP using either (a) the 5-year change in real exchange rate or (b) the spot-rate deviation from OECD PPP fair value. Rank cross-sectionally. Long the 3–5 most undervalued, short the 3–5 most overvalued. Rebalance monthly or quarterly.
• Reported. Standalone Sharpe of ~0.3–0.5 in-sample (Asness et al. report 0.42 for FX value). Kroencke et al. show value + carry + momentum composite lifts a 60/40 portfolio's Sharpe from 0.64 to 1.05.
• Data. Yes. CPI data is free (OECD, BIS); spot rates retail-accessible. OECD PPP figures are public.
• Honest net Sharpe. 0.15–0.30 standalone. Verdad reports just 0.14 even pre-cost from 2000–2010, and negative since 2010. PPP reversion is real on 5–10 year horizons but provides almost no information at the monthly frequency a systematic trader operates on. Quantpedia's implementation reports 7.82% return / 0.36 Sharpe / -39% max drawdown — i.e. the drawdown:Sharpe ratio is terrible.
• Failure mode. Long holding period mismatched to typical capital horizon. Drawdowns when value is wrong (a "cheap" currency keeps getting cheaper) can be 30%+ and last 3–5 years. Strongly negatively correlated to momentum, so the two together are better than either alone.

4. Volatility-Managed Carry (Volatility Scaling)

• Paper. Moreira, Muir (2017), Volatility-Managed Portfolios, Journal of Finance 72(4), 1611–1644. SSRN 2659431 | NBER WP 22208.
• Mechanic. Take any base strategy (carry, momentum, etc.). Each month, scale the gross exposure by target_vol / realized_vol where realized vol is the 1-month or 3-month trailing standard deviation. When vol spikes, you cut size; when vol is dormant, you lever up. Mechanically simple — one line of code on top of an existing signal.
• Reported. Moreira–Muir report Sharpe improvements of +0.2 to +0.5 across asset classes including the currency carry factor. The base carry Sharpe ~0.5 becomes ~0.8–1.0 vol-managed in-sample (1983–2014).
• Data. Yes. Only requires the returns of the underlying strategy.
• Honest net Sharpe. The Moreira–Muir result is one of the most contested in modern asset pricing. Cederburg, O'Doherty, Wang, Yang (2020, JFE) show that vol-managed portfolios do not outperform their unmanaged counterparts out of sample once you account for the in-sample fit of the scaling parameter. DeMiguel et al. (2024, JF) reaches similar conclusions. Realistic net Sharpe lift: 0 to +0.15. The effect is real on momentum (where vol predicts crashes) but mostly a statistical artifact on other factors.
• Failure mode. The scaling parameter is fitted; live-out-of-sample it can hurt. Works least well on slow-moving signals like carry (where vol is already a poor predictor of crashes).

5. Carry + Value + Momentum Composite (FX Style Portfolio)

• Paper. Kroencke, Schindler, Schrimpf (2014, Review of Finance, SSRN) and Asness, Moskowitz, Pedersen (2013, JF, AQR).
• Mechanic. Build all three single-factor portfolios independently (carry rank, 6-month momentum rank, 5-year PPP deviation rank). Combine with roughly equal risk-weights (1/3 each, vol-equalized). Rebalance monthly.
• Reported. Composite Sharpe 0.9–1.1 in-sample. Critically, the three factors are weakly correlated to each other (FX value and FX momentum correlate ~-0.4 to -0.5 — Asness et al.), so a combination has materially better risk-adjusted return than the best individual factor.
• Data. Yes — same as the three single factors.
• Honest net Sharpe. 0.3–0.5 post-cost on G10. The combination is the part of this literature most worth taking seriously: every single factor decayed post-2010, but the negative correlation between value and momentum means the composite has held up better than any solo factor. Verdad's Sharpe-by-decade table makes this clear — even when individual factors went negative, the diversification benefit partially survived.
• Failure mode. All three components are themselves crowded. If carry and momentum both fail in the same regime (as in 2014–2015), the composite fails too.

6. Currency Crash / Skewness Risk Premium

• Paper. Brunnermeier, Nagel, Pedersen (2008), Carry Trades and Currency Crashes, NBER Macro Annual 23. NBER WP 14473. Extended by Lettau, Maggiori, Weber (2014), Conditional Risk Premia in Currency Markets and Other Asset Classes, JFE 114(2), 197–225 (SSRN 2228967).
• Mechanic. Two flavors. (a) Hedged carry: run normal carry but buy out-of-the-money put options on the high-yield leg, paid for by selling OTM calls (a "risk reversal") — this caps the left tail at the cost of some carry. (b) Downside-beta sort: rank currencies by their beta to negative equity market moves (DR-CAPM); long low-beta, short high-beta currencies. The premium is for bearing downside risk.
• Reported. Lettau et al. report DR-CAPM-sorted currency portfolios with Sharpes around 0.6–0.8 in-sample (1974–2010). Brunnermeier-style hedged carry trims gross return ~30% but trims crash drawdowns by ~50%.
• Data. (a) requires FX option vol surfaces — only AUDJPY-style majors are retail-tradable via IBKR. (b) requires equity index returns (free) and currency returns (retail).
• Honest net Sharpe. 0.2–0.4 for the downside-beta sort. The options-hedged version is not retail-feasible at scale — FX option spreads at retail are 10–30 bps on top of underlying.
• Failure mode. Downside-beta is just carry-in-disguise (high-yielders are also high-equity-beta). If equity markets crash sustainably without USD strengthening (rare, but possible — e.g. 2020 Q2), the trade fails differently than carry does.

7. Term Structure / Curvy Trades

• Paper. Bakshi, Panayotov (2013), Predictability of Currency Carry Trades and Asset Pricing Implications, JFE 110(1), 139–163 (ScienceDirect). Extended by Dreher, Gräb (2020), From Carry Trades to Curvy Trades, The World Economy 43(3), 758–780 (ECB WP 2149).
• Mechanic. Instead of ranking on the front-end forward discount (standard carry), rank on the curvature of the forward curve — specifically, the difference between long-dated forward discount (e.g. 10Y) and short-dated forward discount (3M). Currencies with the steepest forward curves earn excess returns; flat/inverted curves underperform.
• Reported. Bakshi-Panayotov implied Sharpe 0.7–0.9 in-sample (1990–2010). Dreher-Gräb confirm out-of-sample on different windows.
• Data. Long-dated forward curves are NOT retail-accessible. You can scrape OIS curves from central bank releases but consistent multi-year cross-currency forward points beyond 1Y are an institutional dataset. Workable workaround: use swap rates from FRED/ECB SDW as a proxy.
• Honest net Sharpe. 0.2–0.3 with the swap-proxy workaround, perhaps 0. Genuinely interesting alternative ranking — but the data friction is real.
• Failure mode. Heavy overlap with standard carry in steady-state; the differentiation shows up in regime transitions (curve flattening), which are rare.

8. FX Volatility Risk Premium (Selling Implied Vol)

• Paper. Della Corte, Ramadorai, Sarno (2016), Volatility Risk Premia and Exchange Rate Predictability, JFE 120(1), 21–40 (ScienceDirect). Della Corte, Kozhan, Neuberger (2021), The Cross-Section of Currency Volatility Premia, JFE 139(3), 950–970 (SSRN 2892114).
• Mechanic. Sell 1-month at-the-money straddles (or variance swaps) on a basket of FX pairs, delta-hedge daily. The premium harvested is the positive gap between implied and realized variance. Cross-sectional variant: sort currencies by IV-RV gap and long the smallest gap / short the largest.
• Reported. Della Corte et al. report Sharpes of 1.0–1.5 for vol-selling strategies in-sample. Annualized returns 8–15% on the underlying notional with volatility 8–10%.
• Data. Requires FX option vol surfaces. Retail can access EURUSD/USDJPY/GBPUSD/AUDUSD vol via IBKR (deliverable FX options), but the universe is tiny vs the institutional 30+ pair sets, and spreads are wide. Variance swaps are not retail-accessible at all.
• Honest net Sharpe. 0 to 0.2 retail. The Sharpes in the literature are calculated at institutional execution: tight option bid/asks, no minimum size friction, ability to delta-hedge continuously. A retail trader replicating this gives back 60–80% of the gross edge to costs. Also — this strategy literally blew up in March 2020 and during the 2022 sterling LDI crisis. Strong left tail.
• Failure mode. Short vol = picking up nickels in front of a steamroller. The Sharpe is high in calm regimes and goes to -2.0 during stress. Sizing must be ruthlessly conservative.

9. Macro-Fundamental: Taylor-Rule Forecasting

• Paper. Molodtsova, Papell (2009), Out-of-Sample Exchange Rate Predictability with Taylor Rule Fundamentals, Journal of International Economics 77(2), 167–180 (ScienceDirect).
• Mechanic. For each currency pair vs USD, compute the implied Taylor-rule policy rate for both economies using inflation and output gap. Predict that the currency with the higher implied policy rate will appreciate. Trade the difference in implied policy rates as the signal, monthly, sorted cross-sectionally.
• Reported. Out-of-sample predictability significant at 5% for 11/12 OECD currencies vs USD (1973–2006). Specific Sharpe ratios not headlined — the paper focuses on statistical predictability, not P&L. Independent backtests put it at 0.3–0.6 in-sample.
• Data. Yes — all free (FRED, OECD MEI).
• Honest net Sharpe. 0.1–0.3. The statistical edge is real but small; macro data is monthly with significant reporting lag (you can't trade on output gap data that won't be revised for 2 quarters). When implemented with strict point-in-time discipline the Sharpe drops materially.
• Failure mode. Worked best 1985–2005. Since 2008 the Fed and major central banks have systematically deviated from Taylor-rule paths, breaking the signal during QE/ZIRP and again during 2022–2023 hiking.

10. Global Imbalance Risk Factor

• Paper. Della Corte, Riddiough, Sarno (2016), Currency Premia and Global Imbalances, Review of Financial Studies 29(8), 2161–2193 (SSRN 2280952). Won the Kepos Award for best investments paper at WFA 2013.
• Mechanic. Sort currencies on a composite of (a) net foreign asset position / GDP and (b) share of external liabilities denominated in domestic currency. Net-debtor countries that fund themselves in foreign currency earn a risk premium — go long those, short net-creditor countries. Rebalance annually (data is low-frequency).
• Reported. Sharpe 0.6–0.8 in-sample (1983–2014). Factor is priced in cross-section of currencies and in equities/bonds — strong external validation.
• Data. IMF International Financial Statistics + Lane-Milesi-Ferretti External Wealth of Nations dataset. Free but with 6–12 month lag.
• Honest net Sharpe. 0.2–0.4. This is one of the more underappreciated strategies. Low turnover (annual), uncorrelated to carry/momentum/value, and the macro story is durable.
• Failure mode. Slow-moving — you can't react to crises. Net debtors get hammered in global crises (2008, 2020), so the drawdown profile is carry-trade-like even if the steady-state Sharpe is decent.

Strategies that don't work (skip these)

A. Pure technical patterns (head-and-shoulders, Fibonacci, Elliott waves)

• Park, Irwin (2007), What Do We Know About the Profitability of Technical Analysis?, Journal of Economic Surveys 21(4), 786–826. Surveys 95 studies; finds that simple rules (filter rules, moving averages) showed weak FX edge pre-1995 that has since disappeared. Visual/discretionary patterns (H&S, flags, Elliott) showed no statistically significant edge in any study with proper controls for data snooping.
• Batchelor & Ramyar (2006) explicitly tested Fibonacci retracement levels: no significance once you account for multiple-testing.
• The Lo–Mamaysky–Wang (2000) head-and-shoulders paper that's frequently cited as supporting technicals actually found tiny effect sizes (~1% annualized) that vanish under realistic transaction costs.
• Skip these. The "edge" is in the eye of the chartist and doesn't survive blind reconstruction.

B. Lunar cycles / day-of-week / seasonality

• Dichev, Janes (2003) reported a lunar effect in stock returns but Yuan, Zheng, Zhu (2006, LSE WP) found the effect was fragile under extreme bounds analysis.
• Lo, MacKinlay (1990) showed that most calendar anomalies (Monday effect, January effect) disappear or reverse after publication.
• No credible FX-specific lunar/seasonal effect survives in the literature beyond noise.
• Skip. Even if a 0.1% per-month edge existed, it would be wiped out by bid-ask.

C. Retail-sentiment-only contrarian strategies

• Despite folk wisdom on FX forums, no peer-reviewed paper finds that retail positioning data (FXCM SSI, OANDA orderbook, Myfxbook) generates a tradeable Sharpe after costs. The studies that exist (e.g. on COT reports for futures) find weak effects in specific commodity currencies, swamped by costs.
• Sentiment plus a fundamental anchor (e.g. carry + extreme positioning) is a different question and may have some merit. Sentiment alone, no.
• Skip standalone sentiment strategies. Use sentiment only as a size-modulator on a fundamental signal.

Recommendation: the 3 worth implementing

1. Dollar Carry / Dollar Factor (Lustig–Roussanov–Verdelhan). Different from the carry already tested (it's time-series on the USD, not cross-sectional). All data is free (FRED IP, G10 forwards). Macro conditioning variable (US IP growth) ties it to a regime that's auditable. Lowest data-friction of the high-Sharpe strategies. Expected retail net Sharpe 0.3–0.5.

2. Carry + Value + Momentum Composite (Kroencke–Schindler–Schrimpf / Asness–Moskowitz–Pedersen). The single most-replicated result in FX literature is that the three factors are diversifying (value/momentum correlate ~-0.5). Even if each component is decayed, the composite has held up better than any solo factor. Vol-equalize the three sub-portfolios and trade on G10 only to keep costs sane. Expected retail net Sharpe 0.3–0.5.

3. Global Imbalance Risk Factor (Della Corte–Riddiough–Sarno). Underappreciated, low-turnover (annual rebalance), uncorrelated to the standard three factors. All data is free (IMF IFS, EWN). Slow enough that bid-ask is a non-issue. Adds genuinely orthogonal risk to a portfolio of the above two. Expected retail net Sharpe 0.2–0.4 standalone, but the diversification value when combined is the real prize.

Combined portfolio expectation. A vol-equalized blend of (1), (2), (3) should target a net Sharpe of 0.5–0.7 on G10 with 15-year backtest discipline. That's the realistic ceiling. Anything claiming higher in the literature either uses institutional data, is in-sample-fitted, or relies on EM execution that retail can't replicate.

Explicitly NOT recommended for the next implementation cycle:

• Volatility-managed overlays (replication failures since 2020 are now the consensus)
• FX vol selling (left tail makes it incompatible with retail risk limits)
• Term-structure curvy trades (data friction kills it)
• Anything technical, lunar, or pure-sentiment

Sources

• Menkhoff, Sarno, Schmeling, Schrimpf (2012) — BIS WP 366 PDF | SSRN | ScienceDirect
• Lustig, Roussanov, Verdelhan (2014) — NBER | SSRN | MIT PDF
• Kroencke, Schindler, Schrimpf (2014) — Oxford Academic | SSRN
• Asness, Moskowitz, Pedersen (2013) — Wiley | AQR
• Moreira, Muir (2017) — Wiley | NBER WP 22208 | SSRN
• Cederburg, O'Doherty, Wang, Yang (2020) — JFE [vol-managed failure-to-replicate]
• DeMiguel et al. (2024) — JF / LBS
• Brunnermeier, Nagel, Pedersen (2008) — NBER WP 14473 | Chapter PDF
• Lettau, Maggiori, Weber (2014) — SSRN
• Bakshi, Panayotov (2013) — JFE
• Dreher, Gräb (2020) — ECB WP 2149
• Della Corte, Ramadorai, Sarno (2016) — JFE
• Della Corte, Kozhan, Neuberger (2021) — SSRN
• Molodtsova, Papell (2009) — ScienceDirect | Authors' PDF
• Della Corte, Riddiough, Sarno (2016) — SSRN | Oxford Academic
• Hurst, Ooi, Pedersen (2017) — Yale PDF | SSRN [century-of-evidence trend following]
• BIS WP 405 (Information Flows in FX) — BIS
• Verdad (2023), Understanding Systematic FX Strategies — Verdad
• Quantpedia, Currency Value Factor – PPP Strategy — Quantpedia
• Yuan, Zheng, Zhu (2006), Are Investors Moonstruck? — LSE PDF
• Park, Irwin (2007), What Do We Know About the Profitability of Technical Analysis? — JES 21(4), 786–826