Blog of a professional trader Good_trade . Trading secrets. Teaching. Investments. Cryptocurrency. Training Brokers NASDAQ NYSE
Trading Systems That Work
I looked through Smartlab in the morning. Found a Russian translation of the article Dean Hoffman, which I read once upon a time in English and I remember it, as it corresponds to my worldview on stock speculation and the development of trading systems.
—————————————————————————————————- Trading systems, who work
The power of computers today makes it easy to optimize any trading system, and results «work» those in history will be more than excellent. but, optimized system and good system — these are different things. For example, even if after optimization on history the system gives retroactive signals 20 from 20, it is far from the fact that similar performance will be in the future (neural networks and others).
The main problem with optimization is you, that markets tend to change over time. Low volatility gives way to high. Strong trending markets go flat (no clear trends). The regulated market is going wild. There are many examples.
This leads to the fact, what the market «X» begins to behave like a market «Y», and the market «Y», in its turn, becomes a market «Z», etc. If the system has been optimized for the market «Z», she will probably run into problems in the market like «X»! This problem of most systems, when the rules are optimized for individual stock indices or individual market sectors. Despite their amazing results in the past, all together, in the future, they may represent a very poisonous mix.
And vice versa, let the system be designed for all types of markets, from «A» to «Z». In this case, it doesn't matter if the market «Z» turns into a market «Y» or market «A» begins to behave like a market «P», etc… And let the markets change their properties, as much as they want, it won't be a problem, if a sustainable system has been developed that takes into account ALL types of markets.
Признаки «over-optimized» system:
unrealistic performance (performance) on history,
the system is designed for one market only (or one market sector),
the system uses different rules (algorithms) for different markets,
different rules are used for «buy» And «sell»,
real transaction losses are not taken into account (slippage and commission),
the system uses the capital management method, not using «нормализацию» (like a system, designed for just one tool),
use of static targets in the system, for example $2000 – StopLoss и $5000 — TakeProfit — some markets take hours to achieve these goals, for other weeks (cm. paragraph 2).
An important feature of robust systems is that, that different markets have an equal impact not the overall result. This is achieved, mainly through «нормализации» differences between individual markets. Наример, natural gas can change in price by thousands of dollars per contract during the day, while Eurodollars only change by hundreds of dollars per contract over the same period. Need a way, to normalize and balance these differences.
The reason, according to which it is necessary to do this is as follows: let the system «торгует» under one contract for natural gas and eurodollars. It may turn out like this, what 90% of all profit or loss accounted for only half of the portfolio. And that means, that performance in the future will be highly dependent on only one market. Systems, which do not depend on one market will be of higher quality and more reliable.
In general, a robust system must comply with the following rules:
successful trading in a large number of markets,
successful trading over a long period,
using the same rules for different markets,
the same rules for initiating long and short positions,
real transaction losses are taken into account (slippage and commission),
«normalization»,
there should be no static criteria in the system (fixed «for the money» goals, for example)
The final touch in building a robust system will be its forward testing.: i.e, if the system was created and tested before 2000 of the year, then all the same rules must be tested on history after this date. This will make it possible to avoid retroactive fitting to the result on the history to a greater extent.