I noticed that most backtesting examples use $100K for initial capital, but doesn't it invalidate the test result? Most of us do not have that kind of money, much less committing it in a single strategy, so we will use a much less amount in live trading. The problem is that many position sizing methods use drawdowns in dollar terms from backtesting. If you got a maximum drawdown of, say, $20K with a $100K account in backtesting, then you could use that maximum drawdown value when your account actually had 100K. Does this mean you that should backtest with initial captial that is close to what you have in your account? Thanks!
It really doesn't matter how much capital you use for backtesting, as long as you look at the results in %, instead of dollar value. A 20% drawdown is a 20% drawdown, no matter how big your account is...You can also look at the profit/loss as in points instead of a certain dollar value...Let's say your system makes 17 ES points a month with an 8 pts max. drawdown. Then once you know this, you can come up with position sizing, basicly how much risk you are comfortable with and how much reward you want to make.
Fifteen years ago when I started writing automated trading systems the initial capital field was in a widely used field. Back then stock trading systems were a big deal to try to convert investors to become traders.The primary purpose of initial capital was to show the investor how to compare results with the security as a buy and hold investor or as a trader using the trading program. The idea was to show that the trading program would make more than the buy and hold investor. So if using the initial capital produced a better return with buy and hold than the trading program you would not use the trading program. To my knowledge initial capital served only this function and was the only validation I have ever used it for.
Sure it matters in many cases, if not most. Risking 2% of 5K is not the same as 2% of 100K. Depending on capital a trading method can be a loser or a winner. Just think about it. For example, trading 1 ES contract per 20K versus 5K?. Does drawdown ring some bells?
Sure. But trading 10 contracts with 100K or 1 contract with 10K is the same difference. Once the account value start to mess with the margin, it does count.But we are talking about backtesting, where the account can be whatever. So I don't see the difference between 50K or 100K. That's why I told him to backtest the pointvalue first, without contract numbers...
If you don't see the difference it is only because you have no real backtesting experience as many other noobs here with thousands of posts in chit-chat.
... that's because if you have a system, I don't ask stupid questions like how much money can you make with it, but I ask how many points can you make with itin a day or week. My postcount has nothing to do with your inability to argue, just so you know...
What's "real" backtesting experience? 50k or 100k doesn't matter. % return matters. Out of sample testing matters. Whether your strategy can scale to 50k or 100k matters. How you're modelling fills on your backtests matters. Your KPIs matter. Et cetera. GOSH! You people are gonna turn me into 777...
It doesn't make a big difference in most cases but can. It typically has to do with commissions and the maximum number of positions you need to open. Some examples: A system that needs to open tons of stock positions. You can't do this with a tiny account due to commissions and small profit per trade. A futures trend following system that trades many markets but makes most of its gains on a few markets. You can't open/trade many markets with a tiny account. If you only trade 1 position at a time it will be okay.. if you are using portfolio techniques then it might not work.
+1 Excellent post. In general, win rate and profit factor are functions of account size and trading frictions (commissions, slippage, etc.). I have seen examples of simple systems, the simplest possible maybe, where commission and account size decide between boom or bust. In this blog there is an example of a system that buys on the close and sells at the next open and whose performance heavily depends on account size and commission rate: http://www.priceactionlab.com/Blog/2011/02/who-needs-the-trading-day/ I suppose you can backtest on a per contract or per share basis and then figure out the minimum account size to cover frictions and margin, drawdown, etc. but in general that is not a good practice IMO because it restricts risk and money management possibilities. So yes, I would say you can do that but for very simple stuff, almost trivial. I always use the real value of my account when I backtest.