"Buy the high-yielding currency and collect the swap." It is the oldest and most widely believed idea in FX. We tested it as an 8-currency, 16-year, monthly-rebalanced portfolio. It made money. We still did not adopt it.
What we tested
At the start of each month we rank eight currencies (USD, EUR, JPY, GBP, AUD, NZD, CAD, CHF) by their central bank policy rate, buy the top two and sell the bottom two. Rates come from official statistics, and only rates published before the rebalance date are used. We fixed the number of variants at three, decided in advance (taking 1, 2 or 3 currencies per side; two is the primary). The more variants you test, the more likely one looks good by chance.
Results
Over 16 years and 193 rebalances: +82.0% (+5.12% per year). Trading costs came to only -0.25%, because the book moves once a month.
| Component | Amount (of notional) | What it means |
|---|---|---|
| Interest component | +44.6% | Earned by holding the higher-yielding currency |
| Exchange-rate move | +37.6% | The currencies themselves rising or falling |
| Trading cost | -0.25% | Paid on one rebalance per month |
The breakdown is what matters
Decomposed, the interest component alone was +44.6% — positive on its own. So the mechanism, "holding a higher-yielding currency earns the rate differential", is real over these 16 years and these 8 currencies. In our other tests (combinations of technical indicators, for example) the expectancy was already zero before costs. This is qualitatively different.
| Year | Return |
|---|---|
| 2010 | +14.4% |
| 2011 | +11.0% |
| 2012 | +12.7% |
| 2013 | +3.9% |
| 2014 | +8.7% |
| 2015 | -7.8% |
| 2016 | +8.7% |
| 2017 | +4.2% |
| 2018 | -2.4% |
| 2019 | +2.3% |
| 2020 | +4.6% |
| 2021 | +5.8% |
| 2022 | -2.1% |
| 2023 | +4.7% |
| 2024 | +9.1% |
| 2025 | +1.3% |
| 2026 | +2.9% |
Three reasons we still did not adopt it
1. Shuffling the currency labels does about as well. Randomly permuting which currency is which, hundreds of times, produced a top-5% result that beat the real strategy. This era ran dollar-strong and yen-weak, so even a nonsense ranking tended to land on the right side. We cannot claim the ranking itself carried the information.
2. It loses to the ranking everyone already knows. Simply holding AUD and NZD long against JPY and CHF for the entire period, never re-ranking, did better than reshuffling every month. The monthly re-rank adds nothing. (That fixed pairing was chosen with hindsight, so we cannot adopt it either.)
3. The broker's swap markup eats it. The profit here is received as swap, but what actually lands in a retail account is the interbank differential minus the broker's markup. Our figure: paying more than 2.56% per year in markup wipes out the return. Given that both legs of the position carry a markup, that is not a comfortable margin.
What we are disclosing
The sample runs 2010–2026 and does not include the 2008 carry-trade crash. This strategy is known for a return distribution that "climbs the stairs and takes the elevator down", so a record that has never met a crash should be read with that discount. The strongest stretch also coincided with a historic extreme in the US–Japan rate gap; through the early 2010s, when rates were uniformly low everywhere, it barely functioned.
How to read this
"Carry works" is not wrong. But it is something the market already knows, and a sizeable share of it is taken by the broker markup before it reaches you. Our standard treats "the mechanism is real" and "you can reproduce it in a retail account" as two separate questions.
About the raw data
The tables above aggregate 3 configurations. The pre-aggregation data (CSV) and the harness design notes are also available. The conclusions, every configuration's result, the pass criteria and the reasons for failure are all published on this page. Nothing inconvenient sits behind the gate. It goes to readers who opened an account through this site (account-number match only; no email needed).
This is a record of testing we ran ourselves. It is not an assessment of any individual, organisation or service. All figures are simulations on historical data and do not indicate future results. Trading can produce losses exceeding your deposit. This is not investment advice. Make your own decisions.
Related articles
Do Grid Orders Across a Range Actually Work? 21 Pairs, 16 Years, Survival Tested
We tested evenly spaced limit orders with small take-profits and no stop-loss across 21 pairs. Without liquidation modelled, every pair profits. With it, survival splits 33%–100%. Here are the pairs that survived a shifted start date.
Do Order Blocks, Fair Value Gaps and Liquidity Sweeps Actually Work? 995,550 Trades Tested
We ran the six setups usually presented with a "70–80% win rate" over 995,550 trades with measured spreads, targets fixed at 2–3× the stop. Here is the outcome — and the specific reason these backtests look good.
Do Fibonacci + Elliott Wave + MACD + RSI Actually Win? 198,000 Backtests Across 38 Instruments
We tested the classic "combine four indicators for a high win rate" claim across 38 instruments, 3 timeframes and 2 brokers — 198,000 backtests. All 15 combinations came in under PF 1.0. We also show exactly where an "80% win rate" comes from.