Premium Market Analysis

Trend following

Momentum or Trend-following? Removing the Bias

We remove data-mining bias, increase samples and compare the performance of momentum and trend-following.

Many articles on momentum and trend-following select some lookback period for the indicators that is known to result in good performance without taking into account data-mining bias. As a result, the performance metrics are affected by the multiple comparisons problem and suffer from bias.

For example, the table below shows the performance of momentum and trend-following with 12-months lookback period in SPY ETF. Backtest period is from SPY inception to 11/19/2019.

Parameter MOM12* TF12**
CAGR 9.9% 8.8%
Max. Drawdown -16.2% -19.8%
Sharpe 0.87 0.83

   *Buy if ROC(12) > 0, exit if ROC(12) <0
** Buy if price > SMA(12), exit if price < SMA(12)

From the above table it appears that momentum outperforms trend-following. However, the performance metrics are biased due to selection of the lookback period. Unfortunately, this is a misunderstood or not understood problem.

Increasing the samples size

The sample for SPY is small. We will use DJIA from 01/02/1914 to 11/18/2019 to increase the sample.

Removing selection bias

We will vary the lookback period n for momentum from 1 to 24 and for TF from 2 to 24 (n=1 is price) Then we will average performance. The results are shown below.

Parameter MOMn TFn
avg. CAGR 4.9% 5.0%
avg. Max. Drawdown -50.2% -52.0%
avg. Sharpe 0.36 0.37

As it turns out then, when bias is removed as much as possible, momentum and trend-following have statistically indistinguishable performance. This should be expected.

More importantly

The average maximum drawdown of about -50% is an estimator of the future drawdown. Some people get the impression that about -20% is the expected drawdown from these timing strategies. But future drawdown will depend on speed of corrections and in fast markets these momentum and trend-following indicators have significant lag. Therefore, large drawdown levels can be realized in the future.

In addition, an annualized return of about 5% should be expected in the next 20 to 50 years, rather than the optimistic 8% or more. People should not confuse expectation with realized values. Expectations help in risk management and portfolio allocation, they are not point forecasts.

I f you have any question do not hesitate to contact me.

Charting and backtesting program: Amibroker

Data provider: Norgate Data

Technical and quantitative analysis of major stock indexes and 34 popular ETFs are included in our Weekly Premium Reports. Market signals for position traders are offered by our premium Market Signals service

If you found this article interesting, you may follow this blog via RSS or Email, or in Twitter