Counter-Trend ETF Trading System

In my reading list for this past weekend there was a Counter-Trend Trading article (courtesy of  The Idea Farm).  The article from 361 Capital argues counter-trend systems are frequently overlooked but deserve consideration.

The article by 361 Capital used a 10 day high/low system as an example of counter-trend system.  The model went short when the S&P 500 hit a 10 day high and went long when the S&P 500 hit a 10 day low. Positions were held for 3, 5, and 10 days in their backtests. The model was intended as an example of a counter-trend system and not as an optimized system, but it did perform well in their backtests.

Counter-trend systems like the one constructed in the paper are intended to have more trades, shorter holding periods, and  a higher percentage of small winning trades. This contrasts to trend-following systems that tend to have fewer winners but much larger returns on the winning trades. In theory, trend following works in trending markets (either up or down), while a counter-trend system should flourish  more in volatile and directionless markets.

Using some of the examples in the paper I created a very simple counter-trend system.  I tested a relative strength system using SPDR S&P 500  (SPY) and ProShares Short S&P 500 (SH).  The ETF with the highest 5 day relative returns was sold short and held for 2 weeks. The closing price the following day was used as the entry price for the short.  This strategy was tested from 2007 until October 26th, 2012.  The results are below and compared to a buy and hold of SPY.  I used return data from ETFReplay.com and then built a custom excel sheet to test the system.

No slippage or commissions were assumed in this test. In addition, no margin costs for a short-sale were assumed. The system used shorting for simplicity sake in the backtesting.  However, given the prevalence of 1x inverse ETFs, shorting is most likely not necessary in many ETF based systems:

The tests show a compound return of 122.3% versus 12.7% of SPY over the same time period. The max drawdown (at each re-balance date) of the counter-trend system was 24.6% versus 52.9% for SPY. However, 26 trades per year in the counter-trend system and any potential tax consequences, margin costs (if any), and slippage would reduce the tests results.  By using the closing price the following day a signal was generated, I attempted to make the system more realistic in terms of trade execution. However, using the closing price of the date the signal was generated presented similar returns.

These tests and results are preliminary and caution is warranted, but I have been experimenting with different timeframes (all of them less than 20 days) and different baskets of ETFs with some mixed success. Relative strength may not be the optimal method to gauge counter-trend; a system based on absolute returns (i.e. a 10 day high or some other time period) for a single security -as opposed to a basket of securities for which relative strength depends–may be more appropriate.  However, I think relative strength counter-trend systems are an under-studied area and at the very least deserve further attention.