It's commercial software. The company who sold it to me (and would be happy to sell it to you) is named at the top of the first page and the bottom of the last page. Look for "LLC" (Limited Liability Company), it goes on the end of company names. Once you type their name into Google, it will find their website for you.
This is a perfect example of "Curve Fitted to death" system. As you might see that it is in the middle of its largest Drawdown at the end of Equity Curve at July 2006. This system does't look worth even the paper its written on.
Curve fitting is a term that really requires some thought before it can have a good or bad meaning. Given enough variables and a genetic algorithm one can create, given a few hundred billion process cycles on a high end PC, an almost perfect trading system. You have fitted the system rules and variables to the data. In fact that is what a neural network or other high end modeling applications do. They find a way to match or learn the measures/variables to the data. The real question is what happens when the system is exposed to data itâs never seen before. If it continues to trade successfully for a spastically significant period of time then the system has discovered factors that predict future market activity. If on the other hand the performance diverges from prior performance then in fact it was curve fitted in a negative sense. The real key is does it work on Out of Sample data? Does it continue to work on going? A good measure of having found a set of parameters that is actually predictive of fundamental market activity as opposed to curve fitting is the distribution characteristic of trades. If the pattern of wins/losses remains similar to the Oust of Sample testing...a few wins, a moderate loss, a few more wins, then you probably have a valid system. If the distribution pattern changes radically it was either curve fitting in the negative sense or the market has changed and the system is loosing its ability to predict future activity. I've seen models trained on say 50,000 bars that continue to have the same trade distribution patterns and profitability characteristics for another 12,000 bars. This indicates that the model has discovered intrinsic market characteristics and is very unlikely to stop working anytime soon.
a sharpe of 1 makes money, a sharpe of 2 is fine, a sharpe of 3.0 spells quiet sleeping all night long.
No, that is the profit factor PF = amount of wins/amount of losses = 7/3 = 2.33 The win/loss ratio = avg win/avg loss I look at PF as the most important parameter. Read carefully the following artcile to understand how the profit factor and win/loss ratio are related to the success rate: http://www.tradingpatterns.com/profitability.pdf The article also discusses how these parameters play in system design and different time frames. A must read. Ron
Ron, Using the PF as parameter , how would you rate a system ? Bad x to x decent x to x good x to x very good x to x exceptional +x In fact the PF is a risk/reward ratio
Bad 0 to 1 (negative profitability) decent 1 to 1.5 (if realized in bear market. Not if prices trend) good 1.5 to 2 (if realized in bear or volatile market. Not if prices trend) very good 2 to 5 (in any market) exceptional +5 For most people, if the wins are 2x the losses that's a very good performance. Ron