Time for a review of a batch of trades I completed over the last 3 months. In this post I summarize the performance statistics, illustrate the actual equity curve and provide a Monte Carlo simulation based on the results in the sample thus far.
The strategy is manually executed, but uses code to generate trading signals and to identify stops and targets. It is a trend-following strategy attempting to enter on pullbacks in price, and is predominantly traded on the 4-hour timeframe in the spot FX markets.
Here are some statistics for the batch of trades, for the numerically inclined:
- 88 trades – by asset class it is 71 in FX, 12 in indices and 5 in commodities
- 48 losing trades and 40 winning trades – producing a win rate of 45.4%
- Avg losing trade -0.634 units of risk , avg winning trade 0.917 units of risk, producing reward to risk ratio of 1.45
- Net Result across 88 trades = profit of 6.23 units of risk – an average of 0.071 units of risk per trade
- Return on amounts risked – for the 88 trades, I risked a total 88 units of risk. Hence a profit of 6.23 implies a 7.1% return on risk.
The equity chart for the strategy looks as follows:
The strategy produced a strong profit over a concentrated space time (see Newsflash: +10R trading week) and then more or less broke even for a while before performance significantly dropped off over the most recent trades. All in all, 88 trades is not a huge sample so not too much should be read into the exact shape of the curve – it will likely be wiser to focus on the average return of 0.07 units of risk per trade.
How was my trading?
All entries, exits and trade management was completed manually. Thus, although I was using code-generated signals for setups, I was effectively still operating as a discretionary trader. However in terms of trading ‘according to my rules’, I did very well. My entry selection was rigid, and so were my trade management and exits – well, most of the time. I didn’t do trading by the seat of my pants and did not pursue trades in the absence of a valid signal. Additionally, I had a lot of processes and routines built into my trading sessions. Thus, all in all, excellent discipline.
What meaning do these results have/give?
Basically close to zero statistical significance. A mere 88 trades is a small sample size. Certainly in the realm of systematic trading, where back-testing regularly runs into the hundreds, if not thousands of trades. It’s likely that the true long-run performance of this strategy cannot be reliably estimated on the basis of this sample. Thus, in a word, pretty meaningless.
It would be more meaningful if I could systematically backtest the strategy and/or build a bigger sample of trades in a simulated practice environment such as ForexTester3.
Let’s assume for a second that the performance to date was actually the true long-run performance. In that case….
Is the performance any good?
I would expect most traders to regard this performance as mediocre at best. That said, consider the following analogy:
The performance is the equivalent of going to any casino’s blackjack table, losing one hand, and getting dealt a blackjack on the next hand – i.e. losing $10, wining $15, losing $10, winning $15 and so on. This doesn’t sound bad at all, does it?
It is fairly easy to play a large number of blackjack hands in a short space of time (100+ hands/hour if the game is reasonably paced) , in other words to get into the long run quickly. It takes far longer to enter the long run taking trades on the basis of code-generated trading signals on the 4-hour timeframe signals in the FX markets.
Let’s use some realistic numbers to get an idea of potential equity curve outcomes over the space of a year. I did 88 trades in roughly a quarter. Ramp this up to 125/qtr and thus 500 trades per year. With that trading frequency and the statistics from above, I ran a simple Monte Carlo simulation for the strategy’s performance. [Thank you to fellow trader Adey for sharing his Excel-based file with me some time ago to enable me to run these simulations!]
The figures in the table on the left indicate the maximum drawdown, max high and so on for both 500 trades (i.e. 1 year) and for 88 trades (the size of my sample) assuming a starting bank of £10,000 and risking £100/trade. The average return over 50 simulations was a whopping 79.9% – but also note that the worst drawdown (DD) in any of those simulations was 58.7%!
One would expect the average return from 500 trades to be 500×0.07 = 35 units of risk or 35% return. The 79.9% average return is achieved via the miracle of compounding. Thus, over the space of a year, a strategy that produces 6.2R over 88 trades and continues to do so, has a lot of wealth-building potential.
Moving onto Part II of the review
Hopefully that was some interesting information for you. In the next post, I will share some of the findings in reviewing the actual trades. The are several objectives with the individual reviews:
- Can the performance of the strategy be improved by altering the setup criteria?
- Can the % of false positives be reduced?
- Can the performance of the strategy be improved by altering the trade management rules?
- Can the strategy parameters be altered to become more suitable for automated, as opposed to manual, back-testing projects?