Time of day: Legit filter or random?

[logic_man]

Dividing up my trades by time of day, I see that two specific hours of the day stand out as negative, whereas all others are positive. Is this something others have found, that their strategies simply don't work during certain parts of the day or is it random? Obviously, I will keep collecting data on it and perhaps it will turn out that over a longer run, the effect will go away, but I'm curious to hear from those who maybe have had a longer time to analyze this effect on their outcomes.

I'm holding one key variable constant in this analysis, so that it is pretty apples to apples. The main thing I think of which could drive it is that the group of traders from whom my strategy takes profits simply aren't in the market during those specific times. Otherwise, it does strike me as arbitrary and not a legitimate rationale for filtering trades.

 

[GTS]

Are you filtering out economic news releases?

 

[logic_man]

Quote from GTS:

Are you filtering out economic news releases?

Not directly, but I have a separate filter for that sort of thing, which seems to work for every other time of day. This seems a bit different than simply that the market moves more around the time of economic news, but I think it is possible that filter is less effective at some times during the day and that could be the source of the breakdown during the times of day I'm looking at, one of which does include a lot of economic news.

Have you, in your experience, simply given up on trading certain times of the day because the results aren't there? Has that decision ever come back to bite you, if you did?

I guess in the worst case scenario of "all traders eventually go bust", at least I'll go bust slower by taking fewer trades.

 

[sheepsucker]

In my research time of day has proven to have some significance, some systems work better at some hours and some at other. But have also unfortunately came to the conclusion that time of day as a parameter can be curvefitted as well as other parameters...

 

[logic_man]

Quote from sheepsucker:

In my research time of day has proven to have some significance, some systems work better at some hours and some at other. But have also unfortunately came to the conclusion that time of day as a parameter can be curvefitted as well as other parameters...

I've also seen research showing time of day effects, so I don't dismiss it offhand, but I do worry about curvefitting. Fortunately, these trades don't make or break the results, so it would just be a matter of potentially missing a few negative expectancy trades along the way.

 

[trade4ever2day]

Random because it cannot have significance as a filter and if it does it is out of pure luck.

 

[RedTankEra]

Not only is the time of the day important but also time of the night.

There are several key tendencies if you willing to do the statistical work of going back one decade and gathering a strong sample.

 

[Laissez Faire]

I don`t think it`s random, but you may have to add other parameters to your analysis in order to be able to use it since not every trading day is the same.

Volatility may on average contract during a certain part of the day, say lunch hour, but on some other type of days that time window will be the most volatile. I suppose it depends on the structure of that particular trading day.

Currently looking into similar ideas, but nothing conclusive yet.

Will keep looking though.

 

[dom993]

I definitively pay attention both to Day of Week & Time of Day ... actually in conjunction.

It is obvious that some time windows are better for some strategies & worse for others.

Over-optimization can be avoided, but for this the filter must have a large enough impact (number of trades) to make it significant in itself - I suggest using a significance test on any filter.

One word of caution though ... if your system uses large reward:risk ratio, filtering in general is a dangerous idea, as it costs your system so much more to miss a win than to take a loss.

 

[logic_man]

Quote from Laissez Faire:

I don`t think it`s random, but you may have to add other parameters to your analysis in order to be able to use it since not every trading day is the same.

Volatility may on average contract during a certain part of the day, say lunch hour, but on some other type of days that time window will be the most volatile. I suppose it depends on the structure of that particular trading day.

Currently looking into similar ideas, but nothing conclusive yet.

Will keep looking though.

Right. The time around 2:15 PM Eastern on a Fed announcement day is going to play out a lot differently than the average day around that time, so I'm not sure that I'd agree time of day is random at all.