A Real Trading Edge, Quantified: Trading and Stop Management

[Desmond_Sterling]

I am reviving this thread for a single reason. I have received a lot of PMs regarding quantitative trading, research and development. To be perfectly honest, i have taken a hiatus from BP. This is due to, put bluntly, the fact I think that this place has degraded in quality of productive and accurate trading conversation, and quantity of good posters. A few BPers whose names I recognized from long ago have msged me regarding my methods and advice. I am bringing this thread back to see if there is any current interest in me continuing to post my research.

I want to put it out there that [B]not all research ends up with something that’s totally profitable, or easily turned into an “EA” or set up[/B]. Its just that something that occurs that may be exploitable, or may not be. The market will test my hypothesis and I just will show the unadulterated, and repeatable result.

If there is interest I will continue, otherwise I will just let this thread die again. As it takes a lot of work and time to produce these types of findings.

I for one am very interested in what you are doing. You might remember me from a few years ago, or, perhaps not. I too took a scientific approach to looking for any kind of trading edge. But while my maths are pretty good, I was lacking the imagination required to investigate the sort of possibilities you’ve done. I’m embarrassed, frankly, because in retrospect what you are showing in this thread should have been… quite basic and obvious as a scientific approach.

A bit about me ~ I’d invested 200 USD into a forex simulator a few years ago, played around with demo accounts and so forth; when I didn’t find an edge I sort of shrugged and focused on other venues for making money. I do electrical engineering consulting; I have a few small businesses ~ forex just sounded quite interesting. I’ve never had a live account. Were I to ever trade, I’d probably code something up based upon a clear statistical edge and let it run. But for me this isn’t so much about money (believe it or not) ~ it’s just plain fascinating to see how it actually all works. Can’t promise I’ll even find this thread in a week or two, but I do check back here from time to time and see how people are doing.

This is a rare precious thread ~ replicable results with theory of operation explained. The scientist in me absolutely wants to hear more. This thread was reason enough to actually dig out my password for this forum and reply.

[peterma]

I know of a trader/educator who uses statistical analysis in his trading and education material, he differs from MeiHua in one aspect - he charges (a lot) for his time.

Many thanks MeiHua, have just re-read the thread through, great stuff.

[MeiHua]

What I am going to show here is the distributions for the time at which highs and lows are put in during the intraday trading session. I am using FXCM data here and they are set to NY time zone. I feel this data is more suitable to this audience for this article rather than my aggregate feed with over 20 liquidity providers. As you will not be able to reproduce it easily without paying for pro data. Also the difference in results are going to be negligible. I also calibrated the charts to roll over at midnight, not any sessions opening. There is a note below.

There is a few things I want to address with this test. This is the first part of a multi part series on the anatomy of a day and possible usage of this information. So be patient. Also I want to address the adage that highs and lows are put in during London and New York. Which is why they are the best times to trade, or so some gurus say. So part my own exploration, part research, and part trading myth busters. Which is exactly the combination of stuff I like to put out.

Testing Procedure:
This test was conducted using 5 minute bars (OHLC) from 6-28-2005 until 1-3-2014. Data was derived from the FXCM feed (not from their platform but from institutional stand alone client). I then use a script to record the times of the highs and lows and then plot in a histogram in excel. its really straight forward.

High Low Distribution over Time (NY)

NOTE: why is there a big spike at midnight? well this comes from the roll over of the script. It is rolling over on 2400 rather than any “session start”. Why did i pick it that way? Because that’s how time is told universally. I read the spike as the start of a unidirectional day, so yesterdays low ends up being today’s high, or vice versa. Although it looks large, its only happened less than 60 times in over 9 years. Don’t let the scale fool you.

Now to the main bit, when and where are the most highs and lows put in? Well if you notice the chart, high and low distribution are basically identical. There are no really large spikes of once without the other, or areas void of any one type.

Lets examine the big chunk between 0000 and 0300. Why is that there? I thought the highs and lows are put in during the London Session? Well it turns out that is false. London session (approx. 0300-0700) really is a lull time of extremes. it seems that the highs and lows are put in during the late Tokyo session. I know this was not expected but hey. its why we do research.

What about that big spike? Turns out that happens at 0700-1000 NY time. Which correlates well to US based news releases like, NFP, Fed report, and the opening of the US markets. So that is actually verified as a extreme time. It also makes highly logical sense, as we all have seen the volatility created by those reports themselves.

Cumulative Distribution of Highs and Lows (NY time)

This chart is much simpler but actually shows more instantly useable information. I am using the 60 % line. Which this graph comes to at around 0925 NY time.

What does it mean? it means that by this time 0925 there is more than 60 % chance that the high or low has been put in for the day. I think everyone can figure out how that is extremely useful. Being able to determine that a high or low will be “set” increases as time goes on. So if your looking for a new high to be put in after say 1600, then your only going to get that < 20 % of the time.

Summary:
Given the first graph. We can see that actually late Tokyo and possibly Tokyo- London cross over are the 1st phase of the markets movements that are going to set an extreme for the day. This is rather different than what I hear floating around the internet about when its going to be placed, in the London session itself. New York Open is actually the absolute best time for an extreme to come into play, and it seems highly correlated to the times of US based news announcements.

The 2nd graph is highly straight forward and easy to understand. After 0925 the chance for a new high or low for the day is less than 40 %, or 60 % that it has already been put in. Its easy to gauge our selves if we are looking for that new high, or new low after that time. We can at least be wary of “asking too much from the market”

[tonyro44]

Thank you sooo much for your hard work, this info is very useful as to make the best use of trading time for maximum results.

@MeiHua

What I am going to show here is the distributions for the time at which highs and lows are put in during the intraday trading session. I am using FXCM data here and they are set to NY time zone. I feel this data is more suitable to this audience for this article rather than my aggregate feed with over 20 liquidity providers. As you will not be able to reproduce it easily without paying for pro data. Also the difference in results are going to be negligible. I also calibrated the charts to roll over at midnight, not any sessions opening. There is a note below.

There is a few things I want to address with this test. This is the first part of a multi part series on the anatomy of a day and possible usage of this information. So be patient. Also I want to address the adage that highs and lows are put in during London and New York. Which is why they are the best times to trade, or so some gurus say. So part my own exploration, part research, and part trading myth busters. Which is exactly the combination of stuff I like to put out.

Testing Procedure:
This test was conducted using 5 minute bars (OHLC) from 6-28-2005 until 1-3-2014. Data was derived from the FXCM feed (not from their platform but from institutional stand alone client). I then use a script to record the times of the highs and lows and then plot in a histogram in excel. its really straight forward.

High Low Distribution over Time (NY)

NOTE: why is there a big spike at midnight? well this comes from the roll over of the script. It is rolling over on 2400 rather than any “session start”. Why did i pick it that way? Because that’s how time is told universally. I read the spike as the start of a unidirectional day, so yesterdays low ends up being today’s high, or vice versa. Although it looks large, its only happened less than 60 times in over 9 years. Don’t let the scale fool you.

Now to the main bit, when and where are the most highs and lows put in? Well if you notice the chart, high and low distribution are basically identical. There are no really large spikes of once without the other, or areas void of any one type.

Lets examine the big chunk between 0000 and 0300. Why is that there? I thought the highs and lows are put in during the London Session? Well it turns out that is false. London session (approx. 0300-0700) really is a lull time of extremes. it seems that the highs and lows are put in during the late Tokyo session. I know this was not expected but hey. its why we do research.

What about that big spike? Turns out that happens at 0700-1000 NY time. Which correlates well to US based news releases like, NFP, Fed report, and the opening of the US markets. So that is actually verified as a extreme time. It also makes highly logical sense, as we all have seen the volatility created by those reports themselves.

Cumulative Distribution of Highs and Lows (NY time)

This chart is much simpler but actually shows more instantly useable information. I am using the 60 % line. Which this graph comes to at around 0925 NY time.

What does it mean? it means that by this time 0925 there is more than 60 % chance that the high or low has been put in for the day. I think everyone can figure out how that is extremely useful. Being able to determine that a high or low will be “set” increases as time goes on. So if your looking for a new high to be put in after say 1600, then your only going to get that < 20 % of the time.

Summary:
Given the first graph. We can see that actually late Tokyo and possibly Tokyo- London cross over are the 1st phase of the markets movements that are going to set an extreme for the day. This is rather different than what I hear floating around the internet about when its going to be placed, in the London session itself. New York Open is actually the absolute best time for an extreme to come into play, and it seems highly correlated to the times of US based news announcements.

The 2nd graph is highly straight forward and easy to understand. After 0925 the chance for a new high or low for the day is less than 40 %, or 60 % that it has already been put in. Its easy to gauge our selves if we are looking for that new high, or new low after that time. We can at least be wary of “asking too much from the market”

[green_as_grass]

If there is interest I will continue, otherwise I will just let this thread die again. As it takes a lot of work and time to produce these types of findings.

Meihua,

I’m glad you are continueing with this thread. Thankyou! I just found it today & appreciate it a lot.

Hopefilly the mods will keep a good troll patrol going so it’s not ruined.

Cheers!

[Kevin_LaCoste]

This is gold. Thank you MeiHua

[o990l6mh]

This is quite interesting.What surprises me the most is the relatively even time distribution of highs and lows. The fact that the Asian session is responsible for a high proportion of them is something I’ve noticed subconsciously but never reflected over.

This begs the thought that even after the Asian session price will have made its high or low to a relatively high degree which should allow for a profitable method even with a win rate of substantially less than 50%.

This type of statistics would be quite interesting to see also for whole week periods as well. Tuesday’s are for instance famously known as Reversal Tuesday, if that’s in fact true a disproportionate amount of weekly lows/highs should be recorded on Tuesdays.

Thank you very much for sharing your research!

[MeiHua]

This is quite interesting.What surprises me the most is the relatively even time distribution of highs and lows. The fact that the Asian session is responsible for a high proportion of them is something I’ve noticed subconsciously but never reflected over.

This begs the thought that even after the Asian session price will have made its high or low to a relatively high degree which should allow for a profitable method even with a win rate of substantially less than 50%.

This type of statistics would be quite interesting to see also for whole week periods as well. Tuesday’s are for instance famously known as Reversal Tuesday, if that’s in fact true a disproportionate amount of weekly lows/highs should be recorded on Tuesdays.

Thank you very much for sharing your research!

This is exactly the type of thinking/reflecting i like to see. Most people just take what people say over even what they experience. So even though you have seen a significant amount of highs lows occur in Asia it wasn’t something that was “confirmed” since not many people talk about that as being a more significant session.

Regarding High/Low of week. Yes this could be done as well. I may put that on the back burner until my little intraday anatomy is done. But i almost feel like that would be short term seasonality, which could also work with the week day directional bias (much like what i posted earlier about weekday combinations mapping). There is definitely a lot of interesting things in that field. But I am not sure how many people are holding for weeks to months as that becomes the hold times for exploiting that kind of behavior. Or maybe like some posters earlier said, do this thread my way. In which case, who cares its interesting!

[peterma]

The fact that the Asian session is responsible for a high proportion of them is something I’ve noticed subconsciously but never reflected over.

Some say that the first thing a London trader looks at, even before he gets to his station, is the high and low of the AS.

[MeiHua]

Some say that the first thing a London trader looks at, even before he gets to his station, is the high and low of the AS.

Some say… His favorite color is blur. He was once roundhouse kicked by chuck norris but he stood there with his arms folded.

All we know is, he’s called the Stig.

[Nill]

[QUOTE=“MeiHua;584821”] Some say… His favorite color is blur. He was once roundhouse kicked by chuck norris but he stood there with his arms folded. All we know is, he’s called the Stig.[/QUOTE]

That made my day! Hahaha

[mrchilled]

Regarding High/Low of week. Yes this could be done as well. I may put that on the back burner until my little intraday anatomy is done. But i almost feel like that would be short term seasonality, which could also work with the week day directional bias (much like what i posted earlier about weekday combinations mapping). There is definitely a lot of interesting things in that field. But I am not sure how many people are holding for weeks to months as that becomes the hold times for exploiting that kind of behavior. Or maybe like some posters earlier said, do this thread my way. In which case, who cares its interesting!

Great thread, so thought I’d throw some stats in:

80% of the time - the high or low of the week is done by end of play tuesday
95% of the time - by end of play Wednesday

Based on 10 years of data for gbp/usd. These stats are on an older thread that I quoted from extensive analysis I did last year to prove some ICT facts. Use this nugget together with a strong directional bias and some S&R lines and price action and you have yourself a strategy that will make you money consistently …

[artamon]

Ok so far what I have learn’t from your findings is:

1. Buy/sell on pullbacks
2. Proper execution of R:R 3:1 or more is needed.
   Unless you have a high accuracy rate 80/90% 1:1 is fine.

What other statistical data have you got laying around that I could use.
Great thread hope you continue it…

Got to agree, great thread! And a deserving bump

Cheers,
Art.

[MeiHua]

Alright so just to make this entry a lot shorter, same data, same provider, same time start (midnight), time zone = NY, same pair EURUSD.

Today I am going to show something I call relative position. This is the relative position of the last price, given the high and low we have seen thus far. It is given in a percentage off the low as to make all days and all instruments the same instead of a price based value. Ratios are much easier to compare.

Example: if the high of the day is 1.3100 and the low is 1.3000 and the current price is 1.3056 then the relative position is .56 or 56% off the the low.

This chart is the Average of all relative prices over the time period of the data (10 years about). Think of it like a seasonality chart, but for the intraday prices.

What do I notice right off? Well the lowest level reading comes in at the same time as our NY interest zone from part I, it actually initiates at the exact same time 0705. I did not plan that actually, its just how it is. But what about the Tokyo session? Well that actually gives us our highest reading on this chart. How interesting, although not exactly lining up with the other chart exactly it is definitely confirming our previous findings.

Also there is a light Bullish bias to the days. as we have the linear regression sloping upward. This could be interpreted very loosely since this is such a smoothed chart as all seasonal charts are. But this could be interpreted a lot of different ways, Highs are put in first in Tokyo, then Lows in NY then we retrace back towards the highs to close out. Or more generally, as time goes on the chance we trade from the low back into the middle of the range is very high. BUT that does not mean if we are at a high we will trade higher, it just means we are going to play inside the middle of the range most frequently later in the day. I just want to make that clear.

The last and final thing to notice is that after 1030ish when we get back to the 50% line, we dont see any more spikes. Again this correlates to our earlier study that after that time we no longer are going to make new price extremes, high or lows most likely.

I actually was hoping this study would have some very blatant new and shiny bits to share, but as all research goes the numbers point the way. I dont look to prove something and find data to show its true, or inject bias into my studies in anyway. But instead we end up confirming more of what I posted earlier. Lets hope the next part has something a bit more interesting.

[mrchilled]

@MeiHua

Alright so just to make this entry a lot shorter, same data, same provider, same time start (midnight), time zone = NY, same pair EURUSD.

Today I am going to show something I call relative position. This is the relative position of the last price, given the high and low we have seen thus far. It is given in a percentage off the low as to make all days and all instruments the same instead of a price based value. Ratios are much easier to compare.

Example: if the high of the day is 1.3100 and the low is 1.3000 and the current price is 1.3056 then the relative position is .56 or 56% off the the low.

This chart is the Average of all relative prices over the time period of the data (10 years about). Think of it like a seasonality chart, but for the intraday prices.

What do I notice right off? Well the lowest level reading comes in at the same time as our NY interest zone from part I, it actually initiates at the exact same time 0705. I did not plan that actually, its just how it is. But what about the Tokyo session? Well that actually gives us our highest reading on this chart. How interesting, although not exactly lining up with the other chart exactly it is definitely confirming our previous findings.

Also there is a light Bullish bias to the days. as we have the linear regression sloping upward. This could be interpreted very loosely since this is such a smoothed chart as all seasonal charts are. But this could be interpreted a lot of different ways, Highs are put in first in Tokyo, then Lows in NY then we retrace back towards the highs to close out. Or more generally, as time goes on the chance we trade from the low back into the middle of the range is very high. BUT that does not mean if we are at a high we will trade higher, it just means we are going to play inside the middle of the range most frequently later in the day. I just want to make that clear.

The last and final thing to notice is that after 1030ish when we get back to the 50% line, we dont see any more spikes. Again this correlates to our earlier study that after that time we no longer are going to make new price extremes, high or lows most likely.

I actually was hoping this study would have some very blatant new and shiny bits to share, but as all research goes the numbers point the way. I dont look to prove something and find data to show its true, or inject bias into my studies in anyway. But instead we end up confirming more of what I posted earlier. Lets hope the next part has something a bit more interesting.

I’m finding it hard to digest this chart, I get what you are trying to show but can’t understand why the range is between 0.45 to 0.52 which seems odd. If you have price at 07:00 at 1.30000 and that co-incides with the low of the day, then you get a score of 0 so to average a score in such a tight band over 10 years doesn’t look right … or am I missing something here?

[MeiHua]

I’m finding it hard to digest this chart, I get what you are trying to show but can’t understand why the range is between 0.45 to 0.52 which seems odd. If you have price at 07:00 at 1.30000 and that co-incides with the low of the day, then you get a score of 0 so to average a score in such a tight band over 10 years doesn’t look right … or am I missing something here?

Actually due to the balanced nature of spot Forex’s directional volatility distribution. Is why its so tight. You can look at a post I did on basic data mining some time ago and find the probably for an up or down swing is about 50-50. If we were looking at different product classes like options, futures, equities, it would look significantly different. Consider that a high or low has been placed at probably every interval over this time period, so that would give us the high level of smoothing due to large sample size.

I dont understand what you mean by looks odd?? It came out how it came out.

[peterma]

Some say… His favorite color is blur. He was once roundhouse kicked by chuck norris but he stood there with his arms folded.

All we know is, he’s called the Stig.

Actually one Ben Collins, and color is white.

You’ve missed my point - have a look at yesterday’s and today’s Fibre for an idea - no, never mind.

[o990l6mh]

@MeiHua

Alright so just to make this entry a lot shorter, same data, same provider, same time start (midnight), time zone = NY, same pair EURUSD.

Today I am going to show something I call relative position. This is the relative position of the last price, given the high and low we have seen thus far. It is given in a percentage off the low as to make all days and all instruments the same instead of a price based value. Ratios are much easier to compare.

Example: if the high of the day is 1.3100 and the low is 1.3000 and the current price is 1.3056 then the relative position is .56 or 56% off the the low.

This chart is the Average of all relative prices over the time period of the data (10 years about). Think of it like a seasonality chart, but for the intraday prices.

What do I notice right off? Well the lowest level reading comes in at the same time as our NY interest zone from part I, it actually initiates at the exact same time 0705. I did not plan that actually, its just how it is. But what about the Tokyo session? Well that actually gives us our highest reading on this chart. How interesting, although not exactly lining up with the other chart exactly it is definitely confirming our previous findings.

Also there is a light Bullish bias to the days. as we have the linear regression sloping upward. This could be interpreted very loosely since this is such a smoothed chart as all seasonal charts are. But this could be interpreted a lot of different ways, Highs are put in first in Tokyo, then Lows in NY then we retrace back towards the highs to close out. Or more generally, as time goes on the chance we trade from the low back into the middle of the range is very high. BUT that does not mean if we are at a high we will trade higher, it just means we are going to play inside the middle of the range most frequently later in the day. I just want to make that clear.

The last and final thing to notice is that after 1030ish when we get back to the 50% line, we dont see any more spikes. Again this correlates to our earlier study that after that time we no longer are going to make new price extremes, high or lows most likely.

I actually was hoping this study would have some very blatant new and shiny bits to share, but as all research goes the numbers point the way. I dont look to prove something and find data to show its true, or inject bias into my studies in anyway. But instead we end up confirming more of what I posted earlier. Lets hope the next part has something a bit more interesting.

What’s your thought on the upward sloping linear regression line and the fact that EURUSD has been, largely, uptrending over the last ten years?

[MeiHua]

What’s your thought on the upward sloping linear regression line and the fact that EURUSD has been, largely, uptrending over the last ten years?

Yes, that’s probably a very big part of it, or any sort of large time frame drift is going to show its little head in these small time frames. I have no way to remove that part of the bias without really distorting the information as its such small intraday intervals.

[PipityPopityPo]

HI MeiHua,

I am very pleased to see you back - the information you posted in this thread literally changed my entire outlook on trading. I can’t believe I am finally enjoying profits!

Please keep up the good work - without people as generous as you, many of us would have given up years ago.

Thank you very much.

Steve L