Since it is possible that others are reading the thread, and perhaps being influenced by our opinions; we both failed to make a clear distinction between trading and evaluation.
1) Trading: Market data (the sample space) is not random because it is a byproduct of human activity. If it was random there would be nothing for traders to exploit. Walking through market data, looking for deviations from random behaviour, is all we can do ... either using computers, or with the naked eye (you call deviations from random behaviour inefficiencies). In your case (option trading) the instrument is a derivation so you have to walk through derived data (underlying price, implied volatility etc) looking for inefficiencies in that (mispriced options?) (Technically speaking, trading mispriced options is an arbtitrage trade because it is based on procedural deviations, from the norm, rather than trading deviations i.e. the underlying can exhibit non- random behaviour and the option can have nrb + mispricing). As an aside: The market makers aren't likely to misprice options .... IMO that occurs in high volume options when traders take over the market i.e. your are trading against other traders bid/ask. (I took an interst in option trading many years ago and still retain a general knowledge of the principles .... I find it very useful for stock trading because all of the strategies can be applied, as synthetics, using stocks). 2) Evaluation: Whether obtained by IS/OOS testing, paper trading or live trading we are always faced with the need to evaluate a series of trades. My argument is that the best way to do that is to measure it against a benchmark ( the null hypothesis) which for trading is randomness. The coin flip model is a readily accessible and easily understood random model, and we can describe our trading system in terms of that model. This prepares us for what to expect when trading. Once we return to live trading, however, the value of our model starts to diminish because we are faced with a new day and a new set of (non-random) data. BTW coin flips don't have to be random. You can, and I did, turn it into a biased coin, with different volatility in the synthetic data it produces ..... and no, I am not an advocate of using synthetic data to test our systems, only to test the behaviour of our benchmark. --- In [email protected], "Phsst" <ph...@...> wrote: > > > > > I apologise for my failures as a teacher. > > No need... I was apparently not paying attention earlier. I did not > realize that you were referring to a trading system that randomly > entered trades against the S&P500. > > I will be sure not to attempt to trade a system like that, for as you > say, it is surely no more reliable than the flip of a coin. [:D] > > > > Best Regards, >
