Another Look at Intraday Effects

I thought the best way to study intraday effects was to create a version of MiniMax II that runs on 5 minute bars. Any occasions where intraday price swings stop us out on the day of entry should show up on 5 min bars, and we can compare those trades to the way they are recorded on daily bars.

First, let me explain that an intraday version of a system designed for daily bars in not a panacea to all problems involving daily bars. For one thing, the E-mini's trade 23 1/2 hours per day, and the session actually spans two different calendar days. The start of the session is on the same day as the end of the previous session, which makes it difficult to identify the start of the session. While I was able to program around this for the most part, there are a few trades -- most notably on holidays -- where the session is hard to identify. This can create trades in the performance report that you would not have in actual trading. Nonetheless, this intraday version is more accurate in assessing exits on the day of entry than the standard, daily bar version.

First, the good news. On the E-mini Nasdaq there is not much difference between the intraday and daily bar versions of MiniMax. I found only four trades over the last two and one-half years (the extent of my intraday data) that differed by more than $100 between the two versions. These trades resulted in an over-statement of the performance by the daily bar version of about $4,800. The last such trade was in November 2001. There are 86 trades over this time period, so the daily version is in error on less than 5% of the trades, representing an overstatement of about $56 per trade on average. In my opinion, this difference is not large enough to worry about.

Now for the bad news. On the E-mini S&P, the difference is much larger. I found 10 trades where the difference between the daily bar and intraday bar versions was larger than $100. These 10 trades amounted to an overstatement of performance by the daily bar version of about $13,000. Although only one trade -- the one on 9/25 mentioned above -- occurred since I first starting tracking the system in real time last May, it appears that the risk of these trades occurring in the future is unacceptably high.

Fortunately, there's an easy solution. I looked for profitable parameters using the intraday version and discovered that better results are achieved by using a larger value for the trend filter. When a larger trend is required for entry, there are fewer trades that get stopped out on the day of entry. I then re-optimized the daily bar version using a larger trend filter value. The reason I performed the optimization on the daily bar version is that I can optimize over more data using daily bars, therefore finding more robust parameter values. When I tested the new parameters on the intraday version, I was able to confirm that I got nearly the same result as on daily bars. Specifically, I found only four trades where the difference between the two versions was greater than $100. The net difference for all four trades was less than $400. In other words, I found a set of parameter values for the ES that gives basically the same results on intraday bars as on daily bars.

Moreover, the results are essentially the same as the historical results for the original parameters on daily bars. This means that the historical record is still accurate provided we change parameter values. The real-time tracking results posted on my site ("Profitability Since Release") are unaffected in any case since there haven't been any of these spurious trades since I released the system (with the exception of the trade on 9/25, which is properly accounted for in the posted results).

I recommend using these new parameter values for the ES. The new values are as follows: 0.4, 0.8, 0.4, 0.4, 3.1, 1.5, 1.5. I would expect that using these parameter values will minimize the occurrence of trades where the TradeStation report differs from your actual trading. Over the long term, this should result in better performance than the original parameter values. I don't recommend any changes to the parameters values for the NQ.

Because I know some of my TradeStation customers will be interested in testing this, I'll send an email message to all MiniMax owners to distribute the intraday version of MiniMax II that I created for this study. It will be available in EasyLanguage versions for TS 4, TS 2000i, and TS 6. Let me emphasize that it's not necessary to take your trades from this 5 minute version of MiniMax. It's designed for testing purposes. The new parameter values that I found are specifically designed to work on daily bars with the current version of MiniMax II v2, including the spreadsheet version.

If there's a general lesson to be learned from this study, it's that TradeStation does in fact tend to over-estimate profits on daily bars due to intraday price swings that are not represented on daily price bars. As I hope I've demonstrated, however, it's possible to minimize this problem by carefully selecting your system parameter values with the help of intraday data. If properly chosen, the parameter values should work nearly as well on daily bars as on intraday data.