This isn't the best advice. I had about 3.5 - 4 years of backtesting with 1 hour candles in ES and while forward testing encountered some things I have never seen before in the marketplace. If you are risking a % of your net worth on this it would be best to get 10 years of information and backtest. You can DL a program that lets you use .csv files and just buy the data from a vendor etc. GL
How long to paper trade? Well if you have a system thats working, right away, no? Unless its money you can't lose. It all depends on you, we are all different.
But if you're trading an automated strategy, the emotional aspect was never an issue anyway, whether demo or live. You might loose faith in your automated strategy trading live, and perhaps shut it down, but you're still relying on it to do the trading. The real question in auto trading in demo mode is how much like live is it ? It's not how am I going to emotionally handle trading in live mode.
You didn't answer my two questions. (I don't think you're comprehending the OP. If an auto system is back tested a lot, AND it's bug free AND it is successful for 30 days of paper trading, how much longer do YOU think it's necessary to paper trade before using real money?)
If your method trades off weekly bars, then 1 month is too little. If your method trades off 5 second bars, then 1 day is too much.
Of course I answered the questions. If you disagree, then say something useful, rather than cutting and pasting someone else's questions. And further, lets have the OP ask their own clarifications.
Could it be that there is some confusion between backtesting, out-of-sample testing and paper trading? I personally believe that backtesting should encompass protoyping and optimisation on past data as well as some phase of unoptimised out-of-sample testing. At that point you should be able to see if your strategy holds up or not and whether you can commit to trading it live. However, I also believe that there is a point in running a parallel test between the strategy in backtest mode (on new data) and the live strategy (on so-called "paper trading"). The only point of this phase being to make sure that results are very close - basically a safety check to ensure that the backtested strategy produced results close to real trading (rather a proxy of real trading that is paper trading). Once you are happy with your parallel run (I guess number of trades is more appropriate than test duration), you can move to live trading (possibly at X% of real trading size to run another safety check between real and paper accounts without committing your full size) In any case I would always keep running the new data on the backtest engine alongside the live trading to monitor any divergence. Now - as to how how much time you need for out-of-sample testing... That's an altogether different question
I see. I found my answer in the post time stamped 02-20-10 07:58 PM. Something Useful: Assuming you've done sufficient back testing and debugging of your code, 30 days is enough to start using $. It's unlikely that you'll get profound insights that conflict with the back testing and paper trading in real time on a fully automated system. For example, looking at 'one run' gives less insight than thousands of back tests. Expect to lose sometimes and look at the overall average to determine the effectiveness of the system. There are good recommendations for starting slowly & reviewing the live trading against paper trading here, but only people from the twilight Zone think you should run it until until you're 'routinely profitable. Time is money, and what would be paper profits could be in your bank account.
What I would personally do if I'm happy with in-sample and out-of-sample backtesting is paper trade the system until you have completed a statistical significant number of trades. This may be 3 months or 12 months or more. If paper trading was in line with what you would expect given the results from testing you can proceed to trading it live with real money. At this stage I would trade the system with minimum position size (perhaps 1/4 or 1/2 of full position size) again until you have completed a statistical significant number of trades. Then on to full size, or if you are conservative, continue to scale up slowly. I agree with what TraderZones has said - especially when he said that "later" and "no hurry" are usually good things in trading. Jez - how long for an out-of-sample test? Well that would depend on the timeframe of the system and what your average trade length is. Generally the aim is to test over a long enough timeframe to get a statistically significant number of trades. For my weekly system which has about a 6 month average holding time, each testing period was 3 years. For my shorter term system which has an average trade length of about 1 day, each testing period was 1 year.
I'm having difficulty setting up NinjaTrader. Has anyone else had trouble with this? I have some theoretical ideas for optimum stock selection and may have to resort to building some excel spreadsheets or databases to do some backtesting unless I can figure something out that is easier. So on a statistical basis how many trades do you think would be sufficient for testing a new strategy? I may have misunderstood before but all my strategies are automated. At some point I would like to make them fully computer automated. On shorter timeframes I may not be able to react so quickly.