hello, lots of questions here.
For testing intra-day systems.
Do you purchase tick data and then direct your testing software to compress to the level you desire, eg 15 minutes?
Or do you buy intraday data that comes pre-compressed at 15 minute bars?
Is it a trap top test on 15 minute bars and decide it is no good so you try 50 minute bars... there are quite a few compressions available that seem popular on intra-day data.
What determines your preference for compression? Why 10 minute over 15 minute?
Do you use off the shelf software that is specifically tailored for intra-day data and testing, or are the regular items ok (TS,TR, B!)
What is considered a decent historical period to test over? Is it shorter because the number of bars is greater? Or is the simple rule of more test trades the better still valid?
If a certain concept in logic works for one period of intra-day testing, is it fair to assume that it may not work at all in another period 5 years later? ie, at the 15 minute level, are changes in market characteristic over time more extreme and frequent? (I think some people use constant optimising to deal with this)
Is this all the questions I have about intra-day testing? (I know the answer to this one).
Intra-day is very new to me in system testing... any help would be great.
damian
Basics of Intra-day system testing
A reply to a few questions from the perspective of someone who runs one unusual system on intraday data:
- From my discretionary observations: Many people use 5/15/30/60 minute periods; some discretionary traders try to frontrun these periods or use fib numbers instead.
- I used the longest time period that didnt significantly reduce profits because I dont like to have to watch the screen closely.
- For testing you can use much shorter periods to get your 30 trades but for my own testing I tried to identify periods when the market was behaving differently to minimize the chance that my system "would not work at some time in future" (quiet times, frothy tops, crashes, etc etc). I have been very glad of this in the last couple of months
- a common discretionary approach is to trade in the direction of the trend at the next higher timeframe ... so you don't go long in 5 min until the 30 min trend is already long. You might find that this results in a lower noise trend following if there are a lot of people doing it.
John
- From my discretionary observations: Many people use 5/15/30/60 minute periods; some discretionary traders try to frontrun these periods or use fib numbers instead.
- I used the longest time period that didnt significantly reduce profits because I dont like to have to watch the screen closely.
- For testing you can use much shorter periods to get your 30 trades but for my own testing I tried to identify periods when the market was behaving differently to minimize the chance that my system "would not work at some time in future" (quiet times, frothy tops, crashes, etc etc). I have been very glad of this in the last couple of months
- a common discretionary approach is to trade in the direction of the trend at the next higher timeframe ... so you don't go long in 5 min until the 30 min trend is already long. You might find that this results in a lower noise trend following if there are a lot of people doing it.
John