============ DH; Well, you gave us mostly enough info; frankly i traded /invested NasdaQQQ for years.Sept % looks overly optimistic, but most back testings are that way. If those are days max candle winners/days ?????; you probably have a real good system. If the time frame is shorter than days or weeks , probably not that good because of slippage & comissions.. It should be/work somewhat similiar with QQQ; for sure it would lose less with cash markets. But its your bank account If it was me I would test[trade it cash/QQQ before i ramped it up with leverage in OCT, NOV DEC,JAN [strong trend months usually-but thats probabilities, not predictions]
- It appears 50% of the total results is generated in 6 months, with lengthy periods of time in-between. Will you have the patience to trade that system ? - How much drawdown are you willing to let the system go into, betfore you pull the plug (ie., how much capital are you going to risk on that system) ? How did you decide on that amount ? - How reliable are your backtesting results? Does each fill make sense? (for LMT orders, do you enforce price trading through the limit before assuming a fill ... for STP orders, do you enforce execution price is always at least the stop price - I would strongly suggest adding 1-tick slippage on all stops, and 2-ticks slippage on market orders). In general, is there enough time between entry order sent & entry order execution to believe in the execution (anything under a couple seconds is too suspect). - Do you fully understand the underlying market behavior that gives this system an edge? - Have you looked at performance as a function of Time-of-Day (ToD) / Day-of-Week (DoW) ?
I used to have about 5 systems running concurrently. The issue I had was that the systems were highly correlated and at times I would have multiple signals and have way to much exposure. The systems did well for awhile but I realized it was only a matter of time before they blew up so I pulled the plug in 2006. Sure enough, had I continued running the systems through the crash in 2008, my account would have been decimated. Good luck thanks fan27
Thanks for clearly writing the pnl. I would run this strategy in a portfolio. The next step is to add a volatility filter so that you can have a method to trade less when vol is low. Previous sentence assumes there is a good correlation between market vol and pnl of this system. If above step works fine, it will reduce the number of trades and increase average profit per trade, should also reduce DD, so you should be able to increase the leverage a bit, which will lead to better absolute returns overall. After above step, you should look at either vol based money management or at previous trade result based money management instead of basic fixed fractional. This step needs a lot of caution as its easy to overfit at this step and get fooled by randomness. If your system has a positive expectancy, applying a slightly martingale money management would help smoothen the equity curve and your sharpe would go up and DD will come down. But as I said this is a territory where a lot of caution and experience is necessary. I would listen very carefully to everything dom993 says, very smart guy. Asks very good questions. Good Luck. Keep us posted how system performs in real world. Btw, what kind of system it is - one trade per day - is it like vol breakout kind of system or something similar? Can you shed some light. Maybe I will learn something
The results all were generated in 2008 and 2011 with both years performing particularly well, 2010 was terrible but 2007 and 2012 are not yet complete so i could get more data points but i don't think its worth the time to keep collecting data points. I will have patience to trade the system as long as it operates alongside other systems i am developing. I would never run this solo and actually i am not very happy with it. Also, i will be using same method EricP wrote about in his journal on this site on when to determine when to deactivate the system. I am more interested in learning where i am going wrong as this is my first attempt at automating my discretionary strategies. Hence, all this advise you guys are giving me will help me tremendously when i start going in depth of my next system. I do think October 2008 was somewhat of outlier due to massive volatility. I feel i could start this strategy will capital as low as $20,000. However, i have no method or formula or reason why i come to this amount apart from draw-down based on testing does not seem that large thus far. I don't know what the correct methodology would be to determine account size apart from expected draw-down or risk of ruin scenarios, is there anything i could use for this? The limit orders for entry are pretty basic and the basis behind entry is very very simple. Each fill will be a limit order and i have added at the moment $5 slippage (however i have $5 slippage per round turn so maybe i should be $5 per contract traded just in case). Stop orders will again be limit orders and if stop is touched and not filled then its not big deal as this will work in my favour as stop does not move and stays fixed, hence i can only get i assume 1tick slippage? Is $10 per round-turn for slippage too much considering that all my orders are limit orders?? Keep in mind limit orders are fixed and price would have to trade through me. Currently results are based on $5 slippage per round-turn. Yes the market behavior is nothing complex just based on my observation over the last 5 years of discretionary trading various strategies. It is somewhat of a intraday breakout system. It is completely based on present data and does not draw from data from any other market or indicators. System trades roughly 15 times per month and never more than once per day. No, could you expand on this point please as i don't fully understand and could be something my system really needs. Thank you. Thanks for all your help, all great points and noted in my diary for future reference.
A link to his post would really help (yes, I am that lazy I won't do a search for it) Going with your gut-feel is almost a guarantee for failure. This is just as important as your system performance, if you under-estimate what normal drawdowns can be you'll stop trading the system at the worst time. I don't know if there is one, several, or no correct methodology, however I can tell you what I do: I do a MonteCarlo simulation (50,000 iterations) for at least 1 year worth of trading or 500 trades (whichever is more), using the trade outcome distribution from backtesting. I then use the drawdown corresponding to 95% (sometimes up to 99.999%) confidence level. This is just non-sense to me. I could barely understand using STPLMT for entries (every-time you don't get a fill you just miss a winning trade - by definition, and this to guard you from rare situations of serious slippage), but never for exits - you are guaranteed that one day or the other, the position will keep going against you until it is liquidated by your broker, leaving your account right at zero. ToD/DoW analysis : open your backtest trade log in excel, use 2 columns to identify for each trade the Day-of-Week (1=Monday to 5=Friday) and Time-of-Day (use the Hour() function). Then create a PivotTable to analyse your trades outcome as a function of DoW / ToD. Usually, that leads to interesting findings.
Automate it and run it - at the end of every day, backtest against that day's data, and note the differences between theory and practice. That's my advice, anyway.
I just put in a volatility filter and it has smoothed out my equity curve somewhat and given me increase in net to now: $39,395.00. Total trades has decreased to 485 and average trade win is now $81.23. The profit factor is now sitting at 1.73 now. Basically it has removed trades made under low volatility conditions which overall after 235 trades made slightly negative profit. This is pretty good for the system as means less trades overall and less frustration of sitting through long periods of no gains. Would anyone say this is over-optimization of the system? Will try the time of day and day of week filter to see if i can find anything. Thanks for all the help guys.
I just tinkered with the volatility filter again and it seems i can curve fit the data very easily if i wanted too. However, i tried to make it as simple and least optimized as possible - if i take the VIX Open and remove any trades made when VIX open was between 20.26 to 16.68 i get a good result. Anything under above and over this range is kept as part of the plan. Therefore, total trades removed was: 204 My net becomes now becomes: Total $42,916.20 avg $83.01 N 517 Net Gains $95,385.60 Net Loss $52,469.40 Profit Factor 1.82 Confidence level: 99.8% And the equity curve is smoother and less time spent doing nothing (please see attachment). My question now has to be is this now far to optimized? I mean what does vix 20.26 have to have any significance as next big winner may be when vix is at 20.25. Also, it includes trades where vix is less than 16.68 which is 81 trades and nets $5706.20. I mean why are trades with very low volatility now performing better? I can understand the principle behind low vix levels having weak performance as that is fundamentally sound and as it is breakout system this is fairly obvious. However, i cannot fathom or come to a conclusion or reasoning behind why vix under 16.68 the trades perform well. Is this over optimized?? Should i simply take all trades below vix 20 out and figure the 81 trades was simply statistically insignificant.
- Are you using intraday readings of the VIX for filtering? Else, make sure you are using the prior-day VIX reading at the close. - The levels you picked are off-course very much optimized ... you should look into using a stochastic on the VIX, a pretty unoptimized lookback period for it as 100 or 200 days, and standard readings such as 20% / 80% (or 10% / 90% - or 15% / 85% ... you get the idea I am sure) - Another standard measure of volatility is the 20-days ATR ... you can try that instead of (or in addition to) the VIX, with the above comment.