Because if you look at the chart for the current method, there is no drawdown on it that is $3k. So the live results are not matching with expectations. Could be an abnormality, but there could be something wrong too.
LOL, lets hope Frosty's account (and his mental stability) can hold out long enough for the big turn around! JJ
The real question is: With days like that, how can you possibly turn the thing on? You got guts, I'll give you that. But as far as the other stuff is concerned. Nope. And facing potential losses like you see here, how can you even think about scaling up in contract size? You either have to learn how to trade real fast, and apply good trading principles to your trading bot, or work with someone who can show you good trading principles and help you redesign your bot (and hope they don't steal stuff from you the way nonlinear5 did). Good luck, Jimmy Jam
I'm speculating that the problem here is bar/candle trading. That may work when there isn't any volatility, but as the volatility goes up, waiting for a bar to form (end of period) before making a decision can be expensive.
Frosty, I had a similar situation several years ago where my auto-trading program killed during backtesting. When I turned it on live (but still paper trading) the results were not as good. What I found was that I had unintenionally written the code to look forward (see the next bar) which made it quite good at determining if it should hold / reverse / or go flat on the current bar. Not having this luxury on real time data obviously there was no way it could match the backtesting results. My point is that backtesting IMHO should be only used to determine if your concept is better than 50/50. After that, turn it on live with sim money and let it run for a while. If the results match your backtesting, you have yourself a winner.
It will be interesting to see how this turns out for Frost in round 2. There's already doubts, which would indicate that sooner or later, this bot will be turned off. And maybe rightfully so given the performance. Frost you seem like an intelligent guy and are trying to find that holy grail of 'set it and forget it' trading. Maybe you need to consider that trading is hard enough as it is, let alone trying to design a computer program to do this for you. There's just too much that changes in the markets from minute-to-minute, day-to-day, month-to-month, etc. to expect one program that was backtested over a very small time scale to perform well over time. It could happen, sure; but this thread is awfully reminiscent of the other thread you had. I wish you the best with your bot, but there may come a time when you need to think about actually trading live with your own brain. I know that is hard if you have a full-time job or other commitments, but this exercise in programming is proving to be a very costly one thus far. Just a few ideas. I know the bot programmers will disagree, and that's fine! It doesn't hurt to get a few different view points.
Precisely. There is another step that can be added. Backtest but write your strategy to run on tick data. I can do that with Tradestation or Multicharts by building data bars in arrays and running the strategy as the current data bar is building, thus emulating realtime activity and with no chance of accessing unknown data at decision time. You can minimize the differences between the tick backtest, simulator, and realtime trading if you take a fill on the backtest only if price goes through your limit order by a tick thus ensuring that you got filled.
I would like to thank everyone for their input thus far. Keep it comming. I have noticed a few posts about backtesting. This is not my first time developing trading systems and I know the common pitfalls that can occur in backtesting. I am sure I covered this already in this thread but will cover it again. The way my backtesting engine was developed there is NO way to see future data. When backtesting off of the captured tick data I have, it is read into the system in order as it came in and the decision tree is executed after each piece. Therefore decisions are made on ONLY data that was truly available at that period in time, because the unavailable data has not been read in yet. Also, one posted suggested running on a sim account to see if results match backtested results. That is precisely what I have been doing for the past 6 months since I turned the system off last time. The backtested and sim results all matched up.
Frosty, have you analyzed your bot results's losing days to see if they are in a pattern? Did it do well the week of Memorial Day? How about the Monday before fed announcements and the fed announcement Tuesdays? I have a bot and it is not probably going to do well this upcoming week so I'm running it against the sim account only.