For quite some time, long-short has been my preferred strategy for trading as well as testing. In tune with previously tried MR strategy. This is a strategy to replicate NIFTY returns with lower volatility.
The strategy is to go long the week's 5 biggest losers in NIFTY and short the 5 biggest gainers. As can be seen, the underlying assumption is that stock returns are mean reverting; which I know from previous tests is only valid for developed markets like US and Europe and does not work in India, Russia or China. The time period of a week was chosen to allow the market to price in any fundamental expectation change. The constructed index is a equi-weighted index. Tests on market-weighted and dollar weighted indices did not yield significant results.
The long leg of the trade significantly outperforms the NIFTY over a 5 year period while the short leg has a negative return. Explanation: NIFTY consistently went up in the period which is why mean reversion strategies at their most basic do not work in developing markets.
The combined strategy has tracked the market well, mostly due to out-performance during last year's crash when the short leg gave significant returns. The return statistics are as follows:
The strategy shows a much higher sharpe ratio than the NIFTY and significantly lower volatility. The long leg of the trade out-performs the markets and has an even higher sharpe ratio with significantly higher volatility.
Without trading filters, I would not recommend this strategy. I implemented a learning algorithm to trade the strategy using PCR as a sentiment indicator which resulted in negative returns for the strategy. I did a similar study some time back to test market bias as a leading indicator/sentiment indicator which had a R-square of 0.2 implying that it was useless.
If I had a good leading indicator, how lovely would it be. To develop it, I had to fall back to macro-economics; market variables with time lags. One thing came out of all of it: only credit markets have any clue what is going on...
