My problem with backtesting is that the data (historical), do not account for the liquidity. So the backtest for your Algo, will always assume when you want to buy or sell, and the price is on the chart, the txn is successfully executed. In real life, that may not be the case.
Yeah backtesting has its pitfalls, but it's a million times better than testing in the real market and losing money when it could be avoided. Usually you can tell right away from backtesting if a system is clearly bunk or not, it's usually fairly evident from the start.
Another downside to backtesting is the emotional angle, and how strict you are at following your system, cutting trades early, making rash decisions etc.
People have used algorithm auto-trading with robots that demonstrated great results with backtesting, but they all eventually fail
I don't use algos myself but I do use a system. If the system gives me a buy or sell indication then that's all it does, indicate. I still make the decision whether to enter or not based on what the market is actually doing at the time.
However, if I test a new system on the last month of data and it completely bombs then I know that had I traded that system with real money then I'd be on a net loss. That's the point of backtesting, sorting out the clearly rubbish systems from ones that have potential.
No system works forever all the time, but some work better in some conditions than others or over shorter periods, backtesting is helpful at telling you that, there is no holy grail...
I was on a demo account for 18 months learning and practicing a strategy. I've tried backtesting before but found how rigid the charts were and the inability to make decent markings on them somewhat inhibiting.
I always recommend to people who want to get into trading that they go onto a demo account for at least 6 months while they work out how the markets operate and so that they can practice with strategies without losing any money.
Problem is that it takes too long and people would rather put their money in immediately and rely on indicators that only work 20% of the time.
Yeah, good advice. Here's my basic system >
Come up with an 'idea'
Backtest the idea at least over a 100 trades or so
Modify, tweek and repeat until happy.
Forward test the system using live data in a demo account until I've doubled the account.
Use it on my real account.
Seems like a lot of work, but it works for me...