Over the last several months I have been looking into a new potential daytrading strategy. This one looks into more illiquid stocks and therefore I won’t reveal the strategy: It’s simply not enough volume for many traders to trade the same. So why do I bother to publish it? I would like to show the results of my testing and later I will show the actual profits (or losess) from real trading. This is a mean reversion strategy.
I have tested the strategy by “live” quotes for a period of about two months now in june and july. I decided to backtest the strategy using data from 2010 and 2011. From this data I picked the best stocks fitting my criteria. The stocks I picked based on the results from 2010 and 2011 I tested in an “out of sample test” for January through July 2012.
I first made a portfolio of stocks based on the following:
- Volatility: I don’t want to have stocks with high volatility (using daily movement)
- Over the last year the stock must have been “stable”, not showing extended moves one or the other way.
- Some industries I know from experience is very tricky to trade (at least for me). So I excluded about 50% of the stocks just on this criteria.
- The stock must be higher than 15 USD (lower priced stocks are more erratic).
- I only use NYSE stocks (basically because of criteria number 1).
Doing this very crude selection I ended with 465 different stocks. Of these 465 stocks 100 were selected for my portfolio using results from 2010 and 2011(yes, some kind of curve fitting). Additional criterias were based on correllation with SPY, mean reversion ability (this is a mean reversion strategy) and average profit per fill. Doing this there is always the possibility of curve fitting, so therefore I want to have an out of sample test. That is the method I have used previously and it has worked reasonably well.
Here is how the accumulated P/L looks like for 2010 and 2011 using the best 100 stocks out of 465 candidates:
Each position is with 5000 USD, the lowest size I can possibly trade. Slippage and commissions are not included. There should be no slippage on entry because I get “hit” with the order (providing liquidity), and exit is on MOC. Commissions are negligible compared to average gain per fill. In total there is about 13 952 fills over these two years (about 30 fills per day). I like to use the law of big numbers. That’s why we get this smooth equity curve (and perhaps because of curve fitting).
So how did this sample of 100 stocks perform in 2012?
Surprisingly good, if you ask me. Is this too good to be true? Above 35 000 USD in accumulated profits with the lowest possible size. Over the 650 day period 65% of the days are profitable. Another point to make is that the P/L goes up when the general volatility in the market picks up (best month is august 2011). That makes sense: the more energy out there, the more to prey on.
Remember that this is just one of several strategies one needs to trade. You can’t just rely on just one strategy.
Here is the P/L per day over the whole period:
And here is the final graph showing monthly P/L:
I’m not naive so I will be happy if I can get 1/3 of the profits of this simulation. This is theory and I know that many of these fills I won’t get in real life due to error in the quotes and data. No matter which data you use it will be worse in real life. That is a sure thing.
I have already started trading this strategy from 1. august with 3 days in a row with decent profits.
Update 18th of August 2012: I have now traded this strategy every day in August and made about 3 000 USD on minimum size. As expected, theory is a lot better than real life. However, the strategy performs well and every day has been positive except for one day with a small loss. The biggest discrepancy from theory is that about 5% of the stocks are “hard to borrow”, ie. I can’t short them. They have on average better return than those which are easy to borrow.