The huge disadvantages of manual trading system creation

In the retail sector traders usually approach algorithmic trading in the same way as it was approached 20 years ago by systematic traders. Traders who want to trade an algorithm on their accounts usually analyse the market manually in the hope of finding some type of logic which – when automated – leads to the creation of a trading strategy that can be run in a set-and-forget manner. However manual system generation has huge competitive disadvantages, given that today’s technology allows more advanced market players to exploit market inefficiencies much more effectively, giving traders who rely on manual strategy creation an incredible disadvantage. Today I am going to talk about why I consider manual strategy creation to be completely obsolete and why I believe success using such strategies will become more and more uncommon as time goes by.

6111954_f520

 

The strategy creation process almost always follows the same path. A trader analyses price data and realizes there might be a potential way to profit by executing a given logic. This set of rules is coded, back-tested, and the results are used to derive further modifications that are used to refine and improve the trading strategy. Using information from back-testing results – such as maximum favorable/adverse excursions – a trader can make changes to a strategy that can lead to significant improvements in back-testing results. Some traders then perform things such as parameter optimizations and out-of-sample validations (which I also believe to be an exercise in futility, see this post) to further refine their strategy into the final version which will go into live trading. The above process generally takes a lot of effort and ends up with the generation of a single – or at most a small group – of trading strategies.

The manual system generation process has several disadvantages. Let me enumerate those which I consider to be the most important:

  • Time and energy investment: A trader will need to invest a significant amount of time and energy in the development of a trading strategy. Many traders take days or even months to develop a strategy that yields satisfactory results in simulations.
  • Trader dependence: The success/failure of the system generation process will depend on the trader’s experience and ideas. Another trader might completely fail to find a similarly profitable strategy.
  • Potential inability to repeat: A trader may have been able to find one strategy that fulfills his/her criteria for live trading but he/she may be unable to repeat the process to generate additional strategies.
  • Non-systematic approach: There is no strictly systematic process that is followed in strategy creation. The evaluation and creation process depends on each trader and usually follows no strict path. One strategy may have taken a week of chart analysis to develop, while another might have needed 5 years of trading experience. Since results are also trader-dependent, the whole sequence of events that leads to the creation of a strategy is not systematic.

Probably the worst disadvantages from the above are the third and fourth. The reason why a potential inability to repeat the process is so important relates with system failure. During the past few years I have met at least 5 traders who have had strategies that had been live trading for years which simply stopped generating profit and reached drawdown levels that warranted their discarding from live trading. These traders were then faced with the prospect of replacing strategies that had often taken years to develop and fine-tune. They were simply unable to generate the needed changes because they were constrained by their personal ability to generate trading systems. The added pressure to replace a previously profitable live trading candidate with something similar or better in a fast manner even made some of them take bad management decisions (trade a poorer strategy instead, for example). A trader who needs to generate systems manually will eventually face this scenario and this can often lead to very troubling times.

The fourth is also a terrible but no so obvious disadvantage. Why is it so bad that an approach is not systematic if in the end it leads to results that are “usable” under live trading conditions ? The answer relates with the level of statistical bias that is imprinted within a trading strategy and the inability to evaluate this bias accurately. A trader that has been trading a given symbol for say, 10 years, introduces a huge amount of bias into the generation of a trading strategy because he has the innate ability to subconsciously “mine” in a more efficient manner. A trader’s experience with a strategy adds up to the level of statistical biases within the creation process. If a trader has spent 6 months back-testing and tweaking a strategy, the level of bias starts to increase significantly because the trader is effectively adding mining bias with each refining step. As the degrees of freedom of a strategy increase, so does the hidden monster: data mining bias. However the problem is not that the strategy has some level of bias but the problem is that we’re simply unable to quantify it due to the process being unsystematic. You cannot realistically quantify things such as trader screen time, number of back-tests run or number of system variations tried within a manual system generation process.

2014-11-17_7-35-20

 

In the end, algorithmic traders who resort to manual system generation – which are the vast majority of retail traders – end up with systems that are hard to produce, difficult to repeat and whose statistical quality is difficult to assess. Even those traders who do manage to at some point produce working strategies often face an insurmountable obstacle when such strategies fail — and they all do eventually. It is not surprising that manual system generation is often only successful in places where there is enough human capital to make the system generation process varied and robust enough (such as in some hedge funds or banks using this same type of systems — note that I’m not talking about market making algorithmic trading here). Nonetheless, there is nothing that fundamentally prevents a manual system generation process from being successful, only that the above issues make the probability of this quite low.

That said, algorithmic system creation also has a full set of caveats that must be surpassed in order to do it successfully. The automatic creation of strategies, with its ability to generate millions of “perfect” back-testing results given large degrees of freedom, can be plagued by unmeasured data-mining bias and can lead to even faster loss of capital when in the hands of inexperienced retail traders (who are often eager to trade perfect looking back-tests with no regard for bias issues). Some implementations – such as genetic programming – have the potential to generate vast arrays of incredibly historically profitable, yet completely worthless trading strategies. A process for the automatic generation of trading strategies must be able to completely measure mining bias and generate systems that can be considered to be based on real historical inefficiencies within a high confidence interval.

Right now I am significantly committed to the generation of algorithmic trading strategies, which we have been doing with the help of GPU cloud mining at Asirikuy. If you would like to learn more about automatic system generation and how you too can create trading strategies in this manner within a community environment  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)

Print Friendly, PDF & Email
You can leave a response, or trackback from your own site.

2 Responses to “The huge disadvantages of manual trading system creation”

  1. Frank says:

    I hugely disagree with you. The reason why you hardly have a single individual with true strategy development and researching skills sitting at any hedge fund or prop firm is because nobody has screen experience anymore, nobody has actually managed large amounts of risk and has demonstrated that they truly understand micro market dynamics, market opponents, different players in a given asset class and product, and a lot more. Most of today’s “algorithmic trading strategy developers/designers” have a CS background and little to no knowledge of the actual trading world, what really determines supply and demand and causes prices to fluctuate and in what fashion.

    The special danger that I identify is that you can nowadays sit a monkey on a computer to push the start (ENTER) button to run a backtest that kicks off a machine learning algorithm (nowadays anyone who wants to impress talks about machine learning first), optimizes the heck out of whatever you throw at the system and…voila…out comes a nice curve-fitted strategy. 99% of such systems are utter and complete garbage, which is PRECISELY why countless hedge funds and prop firms knock on my and those doors who have actual sell-side investment bank or hedge fund risk management and trading experience. Those firms’ number one need is alpha generating strategies and if you could just let a machine learn and out come numerous strategies that are robust and generate risk-adjusted excess returns then such firms would not want to hire anyone.

    The simple truth is that you will always real trading experience and use such knowledge and insight as basic building block. Even then I have witnessed much higher success rate via manual strategy building vs. machine learning algorithms. I could write books about the huge failures of approaching algorithmic trading via Neutral Networks and Machine Learning is hot much higher up in the ranks when it comes to successful strategy design, testing, and implementation.

    At least we are still far away from letting computers research and design successful trading algorithms, certainly when one is not a James Simons.

  2. Abcogito says:

    Thanks for the article which made me wonder was there ever a paper about the return models of algorithmic traders vs. manual traders.

Leave a Reply

WordPress › Error

There has been a critical error on this website.

Learn more about troubleshooting WordPress.