I had been doing all my system testing using RTH (open outcry, regular trading hours) data. However, I recently started trading using electronic trading. When I switched my data test files from RTH to composite, I reran a 20 year test and found substantially different results using composite data compared to testing with RTH data. I understand why there should be some differences:
1. trades executed at different prices in composite vs RTH
2. lower liquidity (at least when electronic trading was in its early stages) led to higher volatility so that daily highs and lows were farther apart in the composite data. This in turn can lead to trades being entered or exited that would not have been entered or exited using only RTH data.
I am interested in the experiences of others in testing and trading using composite versus RTH data and markets. Have others seen significant differences in testing using composite vs RTH data? Is long term (20 years) testing using composite data sensible if early electronic trading had lower liquidity and higher volatility? Do others agree that early electronic trading had lower liquidity and higher volatility? As volume grew in electronic trading, has liquidity increased and has daily composite volatility decreased compared to RTH only trading? How have others handled long term testing with regards to composite versus RTH data?
Testing With RTH Data vs Composite Data
A very good question. Not a straight forward answer. What I do is use the pit history for backtesting and signal generation, but trade in the electronic where ever the volume allows. That turns out to be most of the markets I trade, except Rough Rice (the softs aren't electronic yet).
It comes down to volume and most of the volume is still in the regular hours. But to the extent the volume increases in the off hours, you have to watch it and consider if you want to exit a position if it hits your stop point. I generally do not put stops in the overnight markets, except for currencies and bonds, because the thin markets jump around.
Of course, the big plus is that in the electronics we get much better and quicker fills and and can see the depth.
It comes down to volume and most of the volume is still in the regular hours. But to the extent the volume increases in the off hours, you have to watch it and consider if you want to exit a position if it hits your stop point. I generally do not put stops in the overnight markets, except for currencies and bonds, because the thin markets jump around.
Of course, the big plus is that in the electronics we get much better and quicker fills and and can see the depth.
an example of an alternative method:
bund switched from a 6pm(gmt) close to a 9pm (gmt) close. there is liquidity in the 6-9pm hours.
but, i continue to trade a session end of 6pm.
my reasoning is not based on volume(is it a symptom or cause of the following?):
the typical(average) movement(absolute net change) of the market over those 3 extra hours, is , to date, less than the average open to close movement of a 1hour bar in normal hours. Note that i look at the absolute net change rather than a high-low volatility measure, as this is a more robust measure where possible thin markets/price spikes are concerned.
1hour bars volatility is under my radar, so i concludedc that i can get away with not using the 6-9pm data. this also gives consistency between the system application pre- and post- trading hours extension date.
in currencies, euro/dollar, say, the average net movement of the market outside of the "pit" hours(i.e. overnight/japan hours) can be quite high. I don't trade this market, but as it is open for so many hours a day, and significant movement can occur even at low volume hours, thenby the above analysis method, i may wish to trade a longer session. but i am not a fan of 24hr mkts/execution.....sleeping is a necessity!
bund switched from a 6pm(gmt) close to a 9pm (gmt) close. there is liquidity in the 6-9pm hours.
but, i continue to trade a session end of 6pm.
my reasoning is not based on volume(is it a symptom or cause of the following?):
the typical(average) movement(absolute net change) of the market over those 3 extra hours, is , to date, less than the average open to close movement of a 1hour bar in normal hours. Note that i look at the absolute net change rather than a high-low volatility measure, as this is a more robust measure where possible thin markets/price spikes are concerned.
1hour bars volatility is under my radar, so i concludedc that i can get away with not using the 6-9pm data. this also gives consistency between the system application pre- and post- trading hours extension date.
in currencies, euro/dollar, say, the average net movement of the market outside of the "pit" hours(i.e. overnight/japan hours) can be quite high. I don't trade this market, but as it is open for so many hours a day, and significant movement can occur even at low volume hours, thenby the above analysis method, i may wish to trade a longer session. but i am not a fan of 24hr mkts/execution.....sleeping is a necessity!