Major differences between Intraday and Daily backtests

[jaym]

Hi,

I have been working on a short term reversal type strategy for highly liquid US equities listed on the NASDAQ and NYSE. The system enters the market with limit orders and exits with limit or market orders depending on which exit signal is triggered.

I started testing this system on historical daily data which yielded incredible results. Then I decided to test it on historical intraday data in order to see how it would look on that data. The periodicity for the calculations and everything else remained exactly the same; the signals were just taken in an intraday time frame. The results were terrible and totally uncharacteristic of the EOD testing in all respects.

The only difference I can think of is that the order of execution may be different as trades will be picked on a first to hit basis on the intraday time scale rather then the alphabetical selection of the EOD testing. In order to account for this I adjusted my position sizing rules so that there was a diverse enough allocation per trade to allow for every trade to be taken even if I received entry signals for the entire portfolio.

For implementation I then tried calculating my entry and exit thresholds as channels and then executing the system once these were violated with market orders. This simply yielded worse results.

Even after this my results are still very different. Does anyone have any ideas?

Have any of you experienced anything like this before?


Jay

[choppystride]

I'm a novice myself but here are some suggestions that I can think of:

As Hiramhon suggested, doing a trade by trade comparison is a good idea.

But before you do it, I wonder how do you size you orders? If you're not sizing each trade with a fixed dollar amount, you may want to do so just for the sake of this comparison. Anything else (i.e. fraction of equity) could be affected by differences in intraday sequencing.

Of course, like you've already stated above, make sure you have enough starting capital to take all trades.

Also, I think another (very likely) cause may be due to the differences between your intraday and eod data. For instance, did you run any checks to ensure that:

- they both have the same set of dates with no duplicates and in the correct order?
- their OHLCV match (to an acceptable level) on a daily basis?

Differences may also be due to how your test calculate your equity statistics. For instance, in the eod test, since it doesn't know when each stock reaches its intraday highs and lows (most likely not all at the same time), the intraday run-up and drawdown numbers it gives cannot be precise if you hold multiple instruments at the same time.

This is especially true for entry and exit bars where your trades are executed intraday (i.e. not at open nor at close)

In general, I would be very careful in interpreting the backtest results of a multi-instruments system where multiple trades occur within the same bar, especially when the same instrument does so.

These are just some of the stuff that I can think of right now. There are probably a lot more. But I hope they help.