Currency Momentum: Rules, Evidence and a Monthly Trading Plan

What the historical evidence shows

Menkhoff, Sarno, Schmeling and Schrimpf studied up to 48 currencies from January 1976 to January 2010. Their one-month formation/one-month holding strategy produced an annualized mean excess return of 9.46% before transaction costs, falling to 3.92% with the full quoted bid–ask spreads. The net-return t-statistic was 2.20. These are long–short portfolio results, not a broker account’s compound growth rate. Other parameter combinations lost money after costs. See Tables 1 and 7 of Currency momentum strategies, Journal of Financial Economics (2012).

The paper uses six ranked portfolios and currency excess returns incorporating interest-rate differences. Its broad currency universe and forward-market accounting differ from the seven-currency spot-price model below. The published return does not establish the profitability of our simplified implementation.

The evidence supports investigating relative strength. It does not establish a current EUR/USD signal or a return target for 2026. We have not run or published a broker-specific backtest of this model.

The idea: relative strength across currencies

A currency can be the strongest member of a weak group. If every non-USD currency falls, the least weak currencies can still occupy the long portfolio. That is an intentional feature of a relative ranking, and it can produce losing long trades.

This differs from our time-series trend-following guide, which evaluates each market against its own history. It also differs from carry trading, which selects using financing conditions. Adding either filter creates a new strategy that needs its own evidence.

The working hypothesis is that relative leadership sometimes persists into the following month. The main opposing scenario is a sharp reversal: yesterday’s weakest currencies rebound while the previous leaders decline. A diversified basket can still lose on both sides simultaneously.

Build comparable prices before ranking

Use EUR, GBP, AUD, NZD, JPY, CHF and CAD against USD. Record the final available midpoint quote before 17:00 America/New_York on each month’s last trading day. Use the same convention for every instrument, handle daylight saving correctly and retain the original bid and ask for execution modelling.

Normalize every series to USD per one unit of foreign currency, called P. For EUR/USD, P is the displayed rate. For USD/JPY, P = 1 ÷ USD/JPY. Apply the same inversion to USD/CHF and USD/CAD. Otherwise a rising USD/JPY would incorrectly make the yen look stronger.

Calculate the completed month’s spot-price score:

Momentum score = 100 × (P_this_month / P_previous_month − 1)

For example, USD/JPY moving from 150 to 147 means the normalized yen score is 100 × (150/147 − 1) = +2.04%. It is not −2%. This is an invented arithmetic example, not a historical observation.

Do not substitute a live, unfinished candle for the completed observation. If any currency lacks valid prices, skip that rebalance and record why; do not silently rank a different universe.

A complete monthly model to test

The following choices are our educational specification. Fix them before inspecting performance.

The targets give 50% long-currency and 50% short-currency exposure: 100% gross notional, not 100% cash invested on each side. If equity is $10,000, target $2,500 per leg. Translate the target notional into broker units at the execution price and round down to the permitted volume step. Margin availability is an additional constraint, not a sizing target.

For a selected long JPY position, the corresponding retail order is sell USD/JPY. A short JPY position requires buying USD/JPY. EUR, GBP, AUD and NZD quoted against USD use the same buy/sell direction as the currency selection. Log both the currency direction and actual order direction to catch inversion mistakes.

Worked ranking and portfolio outcomes

Here is a fictional month-end ranking. Scores are used only to decide positions; they are not the returns those positions subsequently earn.

Favourable next month: suppose the two purchased currencies return +2% and +1%, and the two sold currencies return −1% and 0%. Before execution and financing costs, account P/L is approximately 0.25 × (2 + 1 − (−1) − 0) = +1.00%. If total costs equal 0.12% of account equity, the outcome becomes +0.88%, or $88 on $10,000.

Reversal next month: the purchased currencies fall 2% and 1%, while the sold currencies rise 1% and 2%. The same calculation gives −1.50% gross, or −1.62% after the illustrative cost deduction. Balanced long/short notional does not eliminate market risk.

These examples use simple normalized spot returns and ignore small differences from inverse-pair contract accounting. A real backtest must calculate each broker contract’s exact cash P/L, convert it into account currency and include financing. None of these figures is a reported trade or backtest result.

Costs, exposure and exit discipline

A monthly rebalance has costs even when an instrument keeps the same rank. Its target size may have changed, and overnight financing accrues throughout the holding period. Charge spread only on executed turnover, commission according to the account schedule, slippage on actual fills and swap on the broker’s applicable rollover days. Do not assume short financing is the negative of long financing.

The base model exits by time and rank, with no intramonth stop-loss. That makes comparison clear but leaves exposure to severe moves between rebalances. Keep the initial exercise in a demo account. Halving each target to 12.5% creates a separate 50%-gross exposure scenario; it reduces position size but cannot guarantee a maximum loss.

Before testing, document an account-level suspension threshold. For example, a research run might close all positions if marked-to-market equity falls 5% below its running peak and stay in cash until that run ends. This 5% threshold is an illustrative operational rule, not an optimized setting or guaranteed loss cap. Report its results separately from the unsuspended base model; gaps can overshoot it.

Track concentrations as well as individual tickets. Several currencies can respond similarly to a common shock, and apparently offsetting USD legs do not neutralize every economic exposure. Consult the position-size, pip-value and margin calculators when translating a planned allocation into tradable units.

How to validate the model fairly

Use chronological development and untouched test periods. Define the dates and rules first, save every attempted configuration and keep the final test period closed until the specification is frozen. Repeatedly selecting the best lookback on the same data makes the test optimistic.

Evaluate the seven-currency model independently of the published academic portfolio. Include losing months, maximum drawdown, time below the equity peak, gross and net returns, turnover, financing, exposure and a risk-adjusted measure such as Sharpe ratio. State whether returns are arithmetic averages or compounded growth; they are different quantities.

Stress costs at twice the measured level and delay execution to the next liquid session. Check that the result is not dominated by one currency or a single crisis. Then record a forward demo log using unchanged rules. A small number of monthly observations is not enough to infer a stable edge.

Reject deployment if net results turn negative in the untouched sample, depend on unavailable prices or collapse with plausible execution costs. A favourable historical paper cannot repair a failing implementation.