forex-scalping.com

Seasonality in backtests: make sure your test covers December and August

A backtest over a period that skips certain months might give a misleading picture. FX behaves differently at different times of year, and scalping strategies are often sensitive to that.

August tends to be quieter in Europe, with lower volume and thinner ranges. December, especially the last two weeks, has very thin liquidity and odd moves. Late June and the end of each quarter can have rebalancing flows. And daylight saving changes in spring and autumn shift session times relative to server time for a few weeks.

If your backtest covers, say, January to May only, it never sees summer quiet, year-end thinness or the autumn DST shift. The strategy might look good and then struggle in conditions it was never tested on.

The fix is simple: test over at least a couple of full years, so every month appears more than once. Then look at results by month. If certain months are consistently bad, that's useful to know. You might decide to trade smaller or stop during those periods.

Also check whether a strategy relies on a session time that shifts with DST. If the EA uses server time and the broker's offset changes, part of the year might be trading the wrong window.

Full-year coverage won't make a strategy work, but it tells you more about how it behaves.

Splitting results by month in a spreadsheet is quick. Export the backtest trades, add a month column, and use a pivot table for net R and number of trades per month. Comparing the same month across different years shows whether a weak month is a pattern or just one bad year. With only two or three years, the sample per month is small, so I'd look for big, consistent differences rather than reading too much into small ones. December being weak in every year is meaningful. One bad March isn't.

Market conditions shift across years too, not just months. A test period that covers a year of high volatility and one of low volatility gives a more honest picture than two similar years. Rate cycles, crisis periods and quiet stretches all change how strategies behave. If possible, I'd include at least one period of unusual conditions in the test, so you know roughly how the strategy behaves when things aren't normal. The results don't have to be good in that period, just survivable. A strategy that only survives calm years is a risky thing to put real money behind.

August and December are very different from the rest of the year. A strategy that relies on volatility may struggle in August, while one that trades ranges may suffer in the sharp moves of thin December markets.

A good backtest covers several full years, so each season appears more than once. If the results in certain months are consistently poor, a simple rule to trade less in those months can be more useful than trying to optimise the strategy for them.

Holiday periods within the test deserve a closer look too. Christmas, Easter and US Thanksgiving week can produce unusual results, both good and bad. A strategy that made a lot of money in one holiday week might have just caught a freak move on thin liquidity. I'd check whether the best and worst weeks in the backtest fall on holidays. If they do, removing them and seeing how the rest of the year looks gives a more realistic picture of what to expect most of the time. A list of holiday weeks next to the weekly results makes this quick.

Do others run separate backtests for different times of year and compare? I've started doing this to see whether a strategy's edge is consistent or concentrated in certain months. If the strategy only works from October to March, that's useful to know, and I might only run it during those months. If it works fairly evenly across the year, that's more reassuring. A single backtest result over several years can hide these patterns, since the good months make up for the bad ones in the total. Splitting by quarter is a simple place to start.