Hi all, I've been running a futures trading system now successfully for the last 3 years, with annual returns ranging between 80-120%. Last 6 months have things have turned for the worse though, where it seems like the systems has "stopped" working (I put this in quotation marks as I don't know when it will revert to "normal"), with drawdowns now amounting to almost 30%. I have now added a order placement "circuit breaker" to the system, which, if implemented earlier, would have stopped most of the decline at an early stage, however with the result of flat instead of declining performance, however further decline is now impossible. I'm aware of the fact that most algorithmic trading systems have "expiry dates" so to speak, but does anyone have experience how to cope with situations like these? The situation is especially depressing as I have a $10m institutional investor about to commit, hence I fear that this might now not materialize... Thanks in advance for your thoughts and ideas...
Your reference point should be the testing you did PRIOR to the three year run. Did you see a similar drawdown? If so, then I would suggest you need to hold your nerve. If this is one of several profitable systems running together then no problem, the drawdown is part of the system so you keep trading it. Hopefully not all of the systems are experiencing drawdown at the same time, and overall you are profitable. When the drawdown is "unusual" in a historical context then you have a difficult decision to make. But three years is not an especially long time to evaluate whether the drawdown is unusual. Ideally you tested over 10 years+ in different market conditions. If this is your only system, then personally I would reduce size. Perhaps cut size in half until you recover the drawdown. If the system is so great, then it will not take long to recover, right? Ok, so it's a maybe sub optimal, but what is the goal? Survival of the system or squeezing every last penny for the next few months? Survival enables close-to-optimal performance in the LONG RUN. Reducing size also buys you time, and you can work more efficiently on finding new systems without expending all of your energy worried about the drawdown. Ideally you should be able to identify WHY the system has underperformed. If the loss "makes sense" due to recent market conditions then you can formulate a plan to put inplace. Evaluate whether those specific conditions are typical and are likely to continue. Determine a level where you will stop trading completely if it continues to underperform. My experience is that some systems are cyclical and drawdowns are normal. The right thing to do is often to continue trading. You need to weather the storm. But it depends on the system. That's why in an ideal situation you have more than one system running so that one system underperforming does not derail you. Feel free to forward details by PM and I can look into it further
No holding period longer than a day (i.e. no overnight position), average is typically a few hours. Annualized standard deviation of daily returns ranges between 25-35%.
Seem like you do not really know your edge. If you know why your edge works, you will also know why it have stopped working. If you do not know your edge, you are basically trading a blackbox hoping that it will continue to perform. Dangerous and not long term sustainable. Decreasing size is a good thing until it gets back to normal performance. How long time has it been in draw down and what was the previous largest peak-to-valley draw down?
Thanks for your input. The thing is, I have other trading systems, but the problem is that the clients wants THIS particular system, hence I'm out of options here. Back testing drawdowns of similar size have been observed. Position size are automatically adjusted according to a simplified "Kelly size" criterion based on account NAV, i.e. position size scales accordingly, so yes, survival is key of course. Although it's not possible to connect the underperformance to any specific macro economic event, it's simply a matter of an unfavorable market regime where there is currently insufficient intraday direction and too much intraday volatility. Will PM shortly for more details.
This all depends on how you define "edge"? For me it can mean anything, but in my particular case it means that I'm able to generate a signal time series which has a high degree of predictive power measured through various statistical properties. If edge means something very concrete, like Warren Buffett having 1m followers and buying stocks at discounts, or Virtue Finance which has a latency edge with their synthetic NBBO, then no, I'm afraid mine is less down-to-earth...
So you've had Sharpe Ratios around the 2.5 to 4.5 mark for several years and for the last 6 months you're Sharpe has been about -1. What was your SR in backtest? Have you just been lucky for 3 years, or did you expect that kind of performance? (bearing in mind backtests can be misleading; it should be a 'walk forward' analysis and ideally not be [over]fitted). If you're true annual Sharpe was say 2.0 then the chances of hitting -1 over 6 months are indeed pretty slim (back of the envelope, less than 5%). This strongly suggests the 'true' SR has fallen, though there is still a decent probability it is positive. You're trading pretty quickly. With which instruments? Futures, FX, equity....? How many instruments do you have? Have they all seen the same performance degradation? Do you monitor your slippage (price at mid when signal generated versus fill price)? Has that been the result of the fall in performance? How high has that been before? How high is it now? How high did you expect to see it in backtesting? If its pre cost returns that is responsible I'd certainly expect a system trading this quickly to have a decay in performance and require refitting every few years. How long does your backtest go back? Can you get more data and see what happens if you run an automated fitter over time (using rolling out of sample results)? Does a robust fit show the parameters changing? Can you see what happens if you don't refit and let the system run unchanged, does it show performance degradation at this level?