Crossover Whipsaw in a Range
A systematic trader runs a 20/50 EMA crossover on a large-cap equity ETF. Over the past four months the ETF has oscillated between 198 and 204, a 3% band, closing above the 20-day EMA 19 times and below it 17 times. The 20/50 crossover has generated six signals: five lost an average of 0.9% and one gained 1.2%. The 50-day EMA is flat, and ADX(14) has stayed between 11 and 18 for the whole period. The trader proposes replacing the 20/50 pair with a 10/30 pair, which backtests better over the same four months, and doubling position size on each signal to recover the losses. Before changing anything, the trader wants a structured diagnosis of what went wrong and whether the proposed fix is sound.
1. Diagnose why the 20/50 crossover failed. Distinguish between a broken system and an unsuitable regime.
Read the worked answer
The evidence points to regime, not system failure. A 3% band over four months, a flat 50-day EMA, and ADX between 11 and 18 describe a range-bound, low-directional market. A crossover system is a trend-following tool; in a range it is structurally designed to buy near the top and sell near the bottom.
Six signals in four months, with five small losses and one small gain, is the whipsaw signature: many trades, losses similar in size, no large winner. The 19/17 split of closes around the 20-day EMA confirms price is oscillating around the mean rather than trending. None of this says the 20/50 pair is "wrong"; it says the market did not offer the condition the system needs. The correct response is regime detection or a filter, not a parameter change.
2. Evaluate the proposal to switch to 10/30 and double size. What are the statistical and behavioral problems?
Read the worked answer
Two problems. First, sample size and selection: six signals, or even the handful a 10/30 pair would produce, is far too few to distinguish skill from noise. Choosing the pair with the best four-month backtest is curve fitting to one specific price path. A faster pair will trade more often in a range, not less, so it will likely increase whipsaw frequency even if the backtest looks better.
Second, doubling size after losses is a martingale-style response: it increases risk exactly when the system is in an unfavorable regime. If the range persists, losses compound faster. A defensible process is to keep the parameter set fixed, add a regime filter (for example, require ADX(14) above 20 or a close outside the Donchian band before taking crossovers), and test on out-of-sample data across multiple instruments. Position size should be set by volatility and stop distance, not by a desire to recoup.
3. What would you actually do with this system over the next quarter?
Read the worked answer
Suspend discretionary override and add a mechanical filter: only take long crossovers when ADX(14) is above 20 and the 50-day EMA is rising, or when price closes outside the 198-204 band. Accept that the filter will miss the first part of a trend; that is the cost of avoiding chop.
Keep risk per trade constant (for example, 0.5% of equity) so a string of losses is survivable. Log every signal: regime, entry, exit, maximum adverse excursion and maximum favorable excursion. After 30 or more trades there will be evidence rather than a four-month anecdote. If the ETF remains in the 198-204 band, the honest conclusion is that this instrument offers no trend edge right now and the capital is better deployed elsewhere.