This is off topic but I am giving up autotrading for now. After weeks of backtesting I found, at 6am today, that I had calculated commissions wrong and the strategy was about a break even one!! :eek:
breaking even while paying commissions+spreads is a good accomplishment. maybe you should should spend some more time to see if you can sharpen your edge a bit more.
In an attempt to just try to get my bearings back I ran it on my real account but didnt have it transmit the orders, just print out the trades. It would have made +$115 today.. not a big day eitherway so nothing loss by not being live..... waiting for the market to get back in "rhythm" with what the bot expects..... tomorrow i am unsure if i will let it trade or not.. soon i will get it back.. just need to let a few days ride out until I feel its getting its groove back.... I can tell when its grooving, all the trades it makes go positive pretty quicklyl... when its out of sync every trade seems like the worse thing to do at that moment.... so hopefully i'll be able to tell when its getting back in the groove
IMHO it seems like you are making the same mistake all over again. What's the purpose of having a bot if you are just going to turn it on and off on a whim? You are afraid to loss. The market is just shaking you out before the big move/day. Let the bot run until you make a lot of money or it blows your account. Just my 2 cents!
I have to agree. You are attempting to add another layer of decision making (your own with the emotional baggage attached) on top of an automated system. Doing so invalidates all of your backtesting. If you think that the system tends to have good days/bad days why can't you code that into the system, e.g. 2 or 3 losing trades in a row = stop trading for the day? That way you can properly backtest this theory and see if it makes your system more or less profitable (or at least changes the drawdown)
I wish you the best in your trading, and if the drawdown you are experiencing is normal from your backtesting results isn't now the time to be increasing size rather than waiting for it to make money and then begin trading again, i guess you need confidence in your system, thats all, but it can be nerve racking if you are not swinging with a big bat so to speak. just a thought best wishes
Frost, here is what I think might be happening to your bot. You indicated that the bot makes about 3 round trips per day, and that you use market orders. This makes it 6 market orders per day. Your test period for which you posted your equity curve covers 11 months. That means there were about 1320 market orders executed. During the test period, the bot made $17,000 (I am refering to your pic named lastmeth.png on the first page of this thread). These results were from backtesting and simulated trading. Now, assume a very probable thing: in live trading, your fill will be just by 1 tick worse than it was during simulated trading. Do the math, and all of a sudden, your bot's gains dwindle from $17,000 down to $3,800, which is worse than "buy-and-hold" for the period. And if you allow just 2 ticks worse than simulated fill, your results are down to $9,400 net loss. With the expectancy of just $25 per trade (17,000 / 660), your strategy is highly suceptible to even minor disturbances in the execution. Your newer strategy's equity curve looks better, but a two-tick differential between the expected and the actual fill would still take a huge part of its net gain. I really think this is the most probable explanation as to why your good looking strategy becomes a net loser as soon as you start trading it live (for the second time now). What I suggest you do is take a closer look at the difference between your assumed/simulated fills and the actual fills in live trading.
Nonlinear, i don't understand why you assume frosty hasn't allowed for slippage/ commisions. isn't that standard procedure? if frosty did allow this.. wouldn't that rule out your theory?
Frosty already says he compares his live fills against his backtesting engine run against the same data to make sure that live is matching the backtesting. If there were any problems with the backtesting being overly optimistic in terms of fills, using unfair look-ahead (as someone else suggested) or any other coding errors it would be exposed by this comparison. From the first page: