When Does Your Market Really Move?

[BlueHorseshoe 10]

Lots of daytraders who rely on volatility tend to focus on trading around a particular time of day when they believe that a market is more likely to move signficantly. I have been testing this concept.

 

EUR/USD

 

The graph below shows the average range of the EUR/USD currency pair for each hour in the day, calculated over the past 260 trading days (roughly one year). All times are GMT - remember that the time shown on the x-axis is the end time for the one hour measurement, so the bar labelled '9' represents the average trading range between the hours of 8am-9-am.

 

Suprisingly, the greatest ranges actually appear to be found around the time of the New York open rather than the London open (even though the latter is where the greater trading volumes usually occur).

 

I'll work my way through a number of markets popular with daytraders with these studies, and eventually compile them into a table - if you have any particular requests then let me know.

 

BlueHorseshoe

[jpennybags 11]

Hey Blue...

 

I'm curious as to your opinion of using a smaller sample size... say a running 6-8 day period (rather than data for an entire year). The first thought that crossed my mind when considering the original post was that a shorter sample period may be more useful to a day trader.

 

Not trying to side track your project, but just wanted to raise the question... see what you thought.

[BlueHorseshoe 10]

  jpennybags said:

Hey Blue...

 

I'm curious as to your opinion of using a smaller sample size... say a running 6-8 day period (rather than data for an entire year). The first thought that crossed my mind when considering the original post was that a shorter sample period may be more useful to a day trader.

 

Not trying to side track your project, but just wanted to raise the question... see what you thought.

 

Hi JPennybags,

 

It's not my project as such - it's intended for anyone who might find it useful, so I'm very open to suggestions

 

Given that your answer to the hospital births puzzle is correct, then it's an interesting question . . . I wouldn't expect that a smaller sample size would produce runs of like behaviour (in fact I would guess that the 'optimum' from a small sample size would most likely change in a seemingly random fashion), but I could of course be completely wrong.

 

The information above is only intended to represent very generalised behaviour, and finding a way to make use of it as part of a specific strategy would be a whole different challenge.

 

It would be interesting to see how closely volatility on any given day corresponds to the release of economic data, and also how more gradual drifts in the optimal volatility period relate to shifts in the major fundamental focus.

 

BlueHorseshoe

[TheDude 10]

Looking at the chart, it seems the largest volume is actually 3-4 pm GMT - which is the London 'close' (4pm) as positions are squared/books handed over.

 

The US open has 2 components: Euro traders looking at US news releases, and of course the Americans, where as during the Euro open, only the London crowd are there as the Asians are more interested in their local currency (I guess)

[SIUYA 11]

interesting thanks. ...- devils advocate -

 

this merely shows when the largest range is per hour unit, but does not mean that the largest move occurs here - eg; the market might have been trending up for 5 hours at 20 ticks per hour, but then regularly consolidates at hour 6 for a 30 tick range.

 

Probably great for scalpers, or a good indication that at certain periods of the day taking profits, or looking for reversals, or similar strategies might be worth investigation.

 

Definitely information that might be valuable to improve an existing strategy.

[BlueHorseshoe 10]

  TheDude said:

Looking at the chart, it seems the largest volume is actually 3-4 pm GMT - which is the London 'close' (4pm) as positions are squared/books handed over.

 

The US open has 2 components: Euro traders looking at US news releases, and of course the Americans, where as during the Euro open, only the London crowd are there as the Asians are more interested in their local currency (I guess)

 

Hi Dude,

 

It's interesting how two people read the same bar chart differently! The peak is where you suggest, but I saw this as increased volatility going into the US cash session (the second highest bar is the hour of the open).

 

On a side note, this points immediately to the type of error that can creep into analysis when one enters the process assuming that one already knows the outcome; I assumed that a forex market would be governed by the London open, and therefore used GMT - depending on Britsh Summer Time the US open can correspond to either 13:30 or 14:30 GMT.

 

I don't have the knowledge to be able to support or contradict your assertions about activity around the key openings, but from a common sense point of view they make sense - all of Europe and the US can comfortably trade the US open. Your point about the 'handing over of books' is extremely interesting, and not something I would ever have considered.

 

BlueHorseshoe

Edited January 10, 2013 by BlueHorseshoe

[BlueHorseshoe 10]

  SIUYA said:

interesting thanks. ...- devils advocate -

 

this merely shows when the largest range is per hour unit, but does not mean that the largest move occurs here - eg; the market might have been trending up for 5 hours at 20 ticks per hour, but then regularly consolidates at hour 6 for a 30 tick range.

 

Probably great for scalpers, or a good indication that at certain periods of the day taking profits, or looking for reversals, or similar strategies might be worth investigation.

 

Definitely information that might be valuable to improve an existing strategy.

 

Hi SIUYA,

 

You make a good point in that, were I performing this analysis to trade real money I would be far more careful than I have been with this research. It's intended to provide a starting point for people to potentially challenge their own assumptions about when markets generally behave in a fashion that would benefit them.

 

  Quote

this merely shows when the largest range is per hour unit, but does not mean that the largest move occurs here - eg; the market might have been trending up for 5 hours at 20 ticks per hour, but then regularly consolidates at hour 6 for a 30 tick range.

 

I would beg to differ: it does indicate when the largest hourly moves (on average) occur. The movement over 5 hour periods would be shown by analysing 5hr timeframe data. I have made the assumption that most daytraders will open and close positions within an hour. If I was performing this analysis in a more rigorous manner and relying on it to trade then I would do one or more of the following:

 

1. Determine the average holding period for my strategy and use this as the basis for the calculation rather than hourly data.
    
   
2. Perform this analysis based on 'measured moves' rather than movements within a given timeframe. One way to do this would be to calculate when the largest price swings occur (given some formula for identifying the limits of these moves such as the classic three bar pattern), and perform the calculation using the time periods from which these originate. This would then capture the source of a parabolic 5hr move of the kind you describe without any "swing highs".
   

I would add (especially in light of JPennybag's post) that I would regard something like this only as a minor "filtering" element for a strategy, and not something that I would rely on for any kind of entry or exit signal.

 

BlueHorseshoe

Edited January 10, 2013 by BlueHorseshoe

[jpennybags 11]

Blue…

 

Let me phrase the original question that I brought up in a more nuanced way. A little backfill may be necessary:

 

As I sit at the trading desk watching the market move everyday, I form a mental moving average for different events that occur each day. It may have a look back period of 4-5 days. Each day, that day gets added to the mix and the trailing day gets dropped… much like a moving average. I don't do this with any purpose in mind, its simply something that happens because I have a human brain that keeps track of things. I think we are wired this way as a matter of survival… possibly from the hunter gatherer days, but that subject is for another day. This mental account or moving average runs in the background… barely audible, but it is there.

 

What I've noticed is that there are times when patterns will form (at least to my fuzzy math / mental moving average accounting) that will play out over 3 to 4 days. As an example: End of session buying (selling) may increase in strength over a period of days. These events are short term, the pattern disappears, and it's back to randomness.

 

That was my reason behind the question posed about using a shorter data period. Possibly, with increased granularity in the data, and a shorter look back period, patterns may evolve against the background of what could be termed "normal" randomness. How useful this would be depends on one's methods. Personally, I'm fine with the mental MA running in the background (it's easily ignored).

 

The patterns that I speak of may just be part of a random distribution, much like a coin toss that turns up 9 heads in a row if tossed 100 times… don't know. It seems that there is more involved with price movement during the day than a simple coin toss though… that may just be my fuzzy math working again.

[BlueHorseshoe 10]

  jpennybags said:

Possibly, with increased granularity in the data, and a shorter look back period, patterns may evolve against the background of what could be termed "normal" randomness. How useful this would be depends on one's methods. Personally, I'm fine with the mental MA running in the background (it's easily ignored).

 

Hello,

 

I see what you mean, I think, and it's worth investigating.

 

To work out if this is the case then I think the best thing to use would be something like the Wald-Wolfowitz Runs Test to determine the typical number of consecutive days for which the hour exhibiting the largest range remains unchanged.

 

Unfortunately the programming for this starts to get a bit fiddly because the data points that are being compared are 24 periods apart, but I'll try and work on this and post some results early next week.

 

In the meantime, take a look at the attachment below, where you can see how quickly the data settles towards its cummulative average - about twenty periods and it becomes fairly stable.

 

Cheers,

 

BlueHorseshoe

[BlueHorseshoe 10]

The E-Mini S&P500 (@ES)

 

The same testing principles as before, except that this time the data is the @ES future continuous contract, and everything's calibrated to exchange (CME) time.

 

Note the diminutive bar at 1700 hours - this is because trading ceases at 16:15 so the bar only accounts for 15 minutes of trading. I could have resolved this by factoring the entire price difference to the open of the next bar, but then this would suddenly include weekend gaps which wouldn't make much sense.

 

Up next, a comparison of both the ES and the Euro . . .

 

BlueHorseshoe

[BlueHorseshoe 10]

The Euro and the ES Compared

 

This third chart shows the analysis for the Euro (series 1) and the ES (series 2) plotted together. The euro data has been derived from the E7 contract rather than the spot fx price, and all times are CME exchange time.

 

The range for each hour is plotted as a percentage of the sum of the ranges for all hours in that day. Note that this is not the same as a percentage of the total daily range (as hourly bars exhibit considerable overlap).

 

On average, larger ranges seem to be found in the ES later into a 24 hour period than the Euro.

 

It would be interesting to perform tests to see whether volatility (simplistically measured here as hourly range) in the Euro in any way 'leads' volatility in the ES.

 

BlueHorseshoe