When you expect to live from the trading of algorithmic strategies it becomes obvious that your most important concern will be the length of time in which strategies are expected to remain profitable since this will be the “useful period” in which you will be able to extract gains from your strategy. Since all strategies might eventually fail it is necessary to consider how long a strategy might last, what we can do to make strategies last longer and what level of profitability we need to achieve – at a minimum – to survive in the longer term. Through this post I will attempt to provide some guidance around the question of trading system life and what we know and don’t know about this issue.

Whenever you build a strategy you use certain type of hindsight (analysis of past market behavior) to attempt to attain profits in the future. The corner-stone of all technical trading assumes that the future will look like the past in some way and therefore using the tactics which have been successful under past testing could potentially bring in future results which are above those of a random strategy. The relevant question here is therefore the time length in which a system’s long term characteristics will match or exceed those of our backtested results.

–

The good thing here is that we have precedent regarding system development and trading (back to the late 80s) and we therefore know for certain that survival of algorithmic trading strategies for long term periods (+30 years) is possible. This of course does not guarantee that similar (or even these strategies) will survive in the future but it does hint at the fact that trading strategies can indeed survive to long periods of time of trading, holding true to their expected long term statistical characteristics. The strategies which have survived this long are mainly futures trading, wide market strategies using single parameters on large arrays of instruments (from 5 to 40).

Of course some of our strategies might in fact look like these futures old-timers (like Quimichi and Ayotl) – giving us a higher sense of security regarding their success under future markets – but many other strategies we trade are actually quite different, often tuned to the actual inefficiencies of single instruments. Classical system traders (think people like William Erkhart) would have never developed strategies in this way since they consider the possibility of failing “too high” and the robustness of overall strategies too low when they are built this way, something which could be understood as the daily strategies designed 30 years ago used only EOD data and traded very infrequently making large instrument baskets not only a desirable but necessary characteristic in order to give the strategies statistical validity.

Now we are walking on territory which seems to be pretty much unexplored since single instrument strategies developed in this way have never been run for periods as long as 15 years with strategies being run for more than 10 years being actually quite rare. So how can we know the chances of survival of our strategies ? The fact is that survival time might depend on many different factors but after talking to several algorithmic traders who have been trading systems of this character for a few years (more than 5) a few ideas to estimate for how long our strategies might last do come about.

The first thing that quickly became evident as I talked to more people was that systems which are developed for the lesser time frames clearly die more quickly since the market’s short term behavior is much more prone to random change and coming up with systems that exploit false market inefficiencies is easy. A guy I talked about who did extensive research on low time frame systems told me that he abandoned this because system development for these time frames – especially in forex – is quite difficult and systems do not last for long. However he did mention that he has been using a low time frame system successfully for at least the past 4 years BUT in order to do this successfully he had to establish trading with a true direct currenex broker where absolutely no slippage and requotes existed (transparency on low time frame trading is a must). He also highlighted that the development process was arduous and required very high quality data whose cost was more than 4000 USD.

Now several other people I talked to use systems very similar to those developed within Asirikuy. Systems which trade higher time frames (at least 1 hour or above) specifically designed to tackle inefficiencies in one or several different currency pairs. In this case the duration of strategies seems to be related to two extremely important factors. The first one is the liquidity of the pair being used and the second one is the number of degrees of freedom the strategy has. I was shown a couple of system results which have been giving trades for 5 years in line with their expected long term statistics and shown others which failed after 2 or 3 years (reaching some worst case threshold).

It seems that you can even develop systems which are very optimized and have little out of sample testing provided that they are very simple and have a very reduced number of optimizable parameters. As the number of parameters increases stress tests become more and more necessary to eliminate the possibility of strong curve fitting. Liquidity also plays a very important role since the more liquid a pair is, the more fundamental the inefficiencies within it will be to its behavior. This is clearly the reason why EUR/USD strategies tend to be more fit for survival while strategies based on minors or on illiquid pairs (like the GBP/JPY) tend to easily suffer from the exploitation of false inefficiencies which simply disappear as liquidity increases and they are eliminated.

It is important to remember that in order to survive in the long term we need to trade our strategies profitably up to the point where beign aware of their failure (reaching a worst case scenario) puts us above a certain profit threshold. So for example if the worst case scenario determined through some statistical criteria for a strategy is 20% we would need to make at least 25% on an account so that a 20% loss leaves us at break even and at least 50% so that a worst case loss leaves us at a 20% profit.

To sum it up, the advice to develop systems that last seem to be pretty simple. Develop strategies which have a small number of degrees of freedom on highly liquid instruments on the higher time frames and develop many of these strategies so that if one of them fails the other ones will be able to keep bringing you profits. I was happy to find through my research that several of our Asirikuy systems fulfill most of these conditions and Coatl developed systems are also within these lines. Obviously the great thing about system survival is that only with a few systems that survive through a 5-10 year period the losses of many systems which fail can be recovered, bringing in very large profits to the algorithmic trader using them. Nonetheless it is always important to consider that losses from systems that fail will eventually put the threshold of overall average compounded yearly profit to maximum draw down ratio at about 3:1.

If you would like to learn more about my work in automated trading and how you too can develop your own trading systems based on sound trading tactics please consider joining Asirikuy.com, a website filled with educational videos, trading systems, development and a sound, honest and transparent approach towards automated trading in general . I hope you enjoyed this article ! :o)