Hi, A few days ago I had a meeting with a fund that wants to trade through my automatic trading system when it will be prepared. I'm in the development stage and have to prove to them (and to myself) a few things before real trading will begin. The thing is, they told me that they already meet with 10 people in the last 2-3 years that tried to develop an automatic trading system and failed. Little question for big answers: What are the main reasons for failing in such a challenge? The methods, backtesting, past against future etc'... Thanx a lot, Ronen.
in a forward looking market, back testing models, unfortuantly, generally lack. so the idea, and I've looked at a few of these in detail, is to be able to answer the following question: If your model is failing, how soon will you me made aware that it is a model failure, vs. say expected drawdowns, and once failure has been identified, how soon can it be fixed. just to be made aware of my bias: atm i don't' believe one can take a pure model and put it to work for the long term. jmo --RS
I can only speak about the automated systems that I know. I run a few automated strategies myself, professionally. It also depends on what "failure" means, is it just a profit/loss measurement, is it a stable return measurement, or is it able to handle different markets. In my opinion, some automated trading systems fail because the model ultimately does not reflect the real market correctly. At times, a model would work for 3 months, or 3 years, and then the market changed significantly (different market participants, change in regulation, etc), so that the model won't be able to sustain profitability anymore. At times the previous model can never be modified enough to respond to the changes in market, so a sufficiently new model need to be constructed.
You read a bit less about the loser percentage these days. Not long ago, the popular estimate of 90-95 percent was never disputed at ET. In all likelihood, among Players with automatic trading systems the percentage losers is roughly at par.
One of the reasons I asked the question above is that I find it illogical that people can do well in the market by following simple rules or sophisticated methods that take in account only technical quantified information, and a machine can not achieve the same results. Of course, by updating and maintaining the machine according to the changing market. ?
The obvious question on backtesting appears never to occur to backtesters. Are the raw numbers of Open, Close, High & Low a sufficient data input per day for sequential days? Thats the question. The answer is 'No'.
"Are the raw numbers of Open, Close, High & Low a sufficient data input per day for sequential days. Thats the question? " But what about Open, Close, High, Low & volume by ticks of one minute? Not good enough? I know traders that make a lot of money consistently by looking only on these parameters. Some trade only on Bid & Ask without even knowing the rest of the parameters...
I have managed to do it but it is extremely difficult for a whole host of reasons and it does not surprise me at all the majority of people that attempt it fail. I would anticipate the failure rate is much higher than 90% as you have an additional layer of complexity on top of the difficulties already present in trading a market.
Exactly those reasons I'm interested in. What are they? What should I be aware of? Technical issues can be solved and are taking in account (such as: Stability of the trading system, crash scenarios, bugs, behavior in extreme market conditions, etc...). You think that's what breaks the majority?