What are the parameters of your Geometric Brownian motion model? What formula are you using to calculate the option price?
So, that means there were no consecutive days of loss for #3 and #4 runs! Or, the maximum consecutive days of loss for #3 and #4 runs were just 1 day!
The main parameters are: - historical_volatility=30% - initial_spot_value=50 - risk_free_interest_pct=0 - expiredays=63 (ie. 3 months) - max_trade_days=252 - bars_per_day=702 (new: 780) - nTradeDaysPerYear=252 - dividendPct=0 ... Black-Scholes: the usual input params, ie. - spot=the current spot - strike=system uses ATM, but this can also be a certain delta pct from ATM - expdays=upto 63, cf. above, and: does not trade (and closes any open pos) if < 3.5 days - volapct=30 - interestpct=0 ... It's a standard Block-Scholes formula, you can find it also on the net, wikipedia has also good explanation and maybe also links to implementations. System has some more params, like nBots, Max_MinPosSize, ...
Something isn't right if you are able to algorithmically beat a random price sequence where the expected value should be zero. According to the blacksholes framework, the E() should be zero. You are saying you can earn 1000% / annually by timing the random data where volatility = time decay of your options. This is impossible. There has to be a bug in your system.
This really needs an answer. There should be no way for you to systematically extract profits from random noise. Unless you are imposing some structure on the randomness that your system can exploit. Are you imposing mean-reversion (or fractionality) in the simulated gBm? What about correlations between assets? Some form of stochastic volatility?
Is that the time frame for decisions is based on 1-second bars? What's the maximum days for a trade? 8 days ? 10 days?
Please see my previous posting (params used etc.). Of course there is no way to systematically extract profits. The system temporarily increases its risk, upto a certain level, and then again reduces it... Maybe indirectly it could use mean reversion tactics as it uses statistics and probabilities. As said, correlation analysis between the assets is not done, but a kind of hedging is very well done, ie. risk sharing. Stochastic vola: not that I'm aware of, as said the system doesn't make any use of the parameter volatility, neither the historic one nor the actually observed one.
My point was that in year like 2008-2009, correlation was extreme and most likely your 5% DD target would look more like 35%. You yourself said that your model assumes normal markets and in normal markets, the correlation is much less. It's the harsh market conditions that decide how good the system is. I'm curious, why won't you test your method yourself using a smaller sum of money. As a programmer you probably have solid savings and could attract capital easily outside of ET if you have done your stuff properly.