Monthly Archives: July 2013

Times for extremes in session “waves”

New sessions statistic! Racked my head for a good bit to figure out how to correctly count the hours because the data set was being difficult. Capture

This data is looking at the time, count, and % of high/low within each session. This is suggesting that in the Asian session, for example, the high/low is most likely to be formed during the first hour, or close to the end. This is actually pretty neat, and lines up with my observations of the market. Since we’re looking at extremes and there are 2 a day, it makes sense that one extreme is formed in the beginning and end..IF we’re making the assumption that each session presents it’s own wave, kinda. I believe that waves follow simple price/time through order flow, not bound by session opens and closes. Money moves markets when it’s time to do so, not when it’s a certain time. We’re seeing other things that make sense, such as the frequency of price extremes created (or rather not created) during the NY session. There’s a heavy drop off as the day gets later, as most extremes are made in the first few hours. Even though the data presented splits itself into Asian, London, and NY, they seem to still point at the same thing. Extremes made during the beginnings and ends of sessions, and particularly in the overlap between two sessions.

There are plenty of interesting things to note, one being the following: 8:00 (NY Open) is the extreme of the London Session 9% of the time, while it marks the extreme of the NY session 32% of the time.

Basic Session Waves Statistic

Been a while since I’ve done some fx work. Unfortunately no real break through or new ideas. Here’s some stats I pulled out today that I’ve been half working on in the past week; Yesterday I took out some time to use some excel to properly extract the data points I wanted.

I wanted to look at some statistics regarding the “waves” that happen with respect to the sessions. The “Asian Wave” or “London Wave” I already see a lot of problems conceptually with this model but I thought I’d give it a try anyway. I’ll write those potential problems at the end. But for now, here is what the data looks at. I define the Asian wave as starting at the Asian open and closing at the London Open. Likewise, the London Wave begins at the London Open and closes at the NY Open.

The basic question asked is “how do the waves move in comparison to each other?”

nada

 

The answer is a lot of coin flipping. I define contra as simply a close in the other direction. I just checked to see if the bars closed red or green without regards to magnitude. In more “wave” speak, I checked for the chances of AB, BC, AAA, ABC, ABB, AAB. I get nothin’.

 

Problems: It’s most likely the case that waves are forming in the middle of the sessions rather than right on the open. It would be too predictable to have stats like this show something. Of course, while we do expect London to move in one direction (over all) and NY to move in one direction, there isn’t something supporting NY moving contra to London. Instead what is possible is a retracement in NY contra to London, with a varying magnitude. Also, in looking at charts, I look at tops and bottoms of the sessions rather than opens and closes; this is probably another statistic I can work on.

 

Week Max-Min Statistic

One thing that’s awesome about this journey is learning new things in excel.

my data set was printing dates as year.day.month, ex 1971.02.05. Unfortunately excel can’t read that as a day, so i had to reformat it into 19710205, and then use =date(left(),mid(),right()). I then changed it into the day of the week using =TEXT(A1, “ddd”) so that I could look at full weeks of data. After using a few filters and some crude programming on my part, I got a list!

excerpt

Pivot table from the list:

extreme

 

The expected % should be about 17%, or 100/6. Note that the total of 747 is weeks of data.

If I were to draw this onto a chart it would look something like..

chart

 

Here’s what I’m taking from it. First, it looks like Sundays are uneventful, as expected. Second, the % of a weekly extreme being hit is highest on Friday, and then Monday. Mondays+Fridays make up close to 50%(!) of the weekly extreme frequency. If we start with the Friday statistic, this is suggesting that the week moves in a trend, or a weekly wave. Cool. Close to 30% of the time, the extreme isn’t hit until then, meaning if it’s been moving up for the past few days, there’s a good shot it will continue to do so until sometime on Friday. Next, Mondays seem to be a good time for a trend reversal, either from the week before it or simply from Sunday (This data is from Oanda servers btw, so Sunday data is just from Asian open until mid night NY, or 17:00-00:00 EST). Since this is calculating both highs and lows, it’s open to the possibility that in a “bull weekly wave” the low is made on Monday followed by the high on Friday and vice versa for bear weekly waves? Hmm.. Possibly more statistics coming out of this one later.

Dear Journal #1

What to do?

I think the basis point is to start with extremes and whats happening at those areas. This ties in with the 20 pip theory from the beginning of the thread. I’m having an idea of looking at extreme TCD long-TCD short, but after locating them, how do I “check out” what’s happening in those areas? What kind of PA would suggest something unusual?

In tracking the subordinate directions gain of momentum, what do I need? What can track momentum? In SB terms, what delta will show momentum strength/weakness?

Perhaps I need to spend more time on why and not how. If I had a wave indi, then what? Why do I need one? to map price action, but what next?

 

Another recreation statistic should be rolling out tomorrow.

Candle prediction statistic

Big

(Data used: 1 Hr data)

This is eventually the kind of thing I want to be doing with waves eventually… I think.

There’s a good bit of stuff in this picture that’s less self explanatory than my other statistics. I showed a bit of the actual data set for reference.

What I’m trying to look at here is how many consecutive bars is too many? How often do certain candle patterns show in terms of pure color? In answering a very basic question, if the bar is green, what’s the likely hood that the next bar is green as well?

Null bars should perhaps be labeled as dojis: the open and close are the same.

The first statistic is looking at probability of occurrence without regard to exclusivity. This can be shown by the snap shot of the output on the left and right side. There is a portion where the data is green, green, green, green, green, green. The right side will show that the pattern of a green bar followed by another green bar happens 5x here (as shown by the listing of 2 con for (2 consecutive) 5 times in a row). In practical terms, this means that if we see 2 green bars, the probability that the next bar will be green is still ~44.74% (I hope this is correct at least O_o). The frequency %’s are taken in comparison to the total of 90159 bars. This means we expect to see a green followed by a red or a red followed by a green 55% of the time. We expect to see two greens or two reds together about 44% of the time. The chance you see 7 of 1 color in a row has a chance of .7%

However, I think the second statistic takes away some of the blurriness of the first. This is searching for a very specific pattern; that the consecutive bar color will be then broken. These frequencies are taken with respect to their non exclusive patterns:

fix

For example, the first RG/GR remains the same, but the second now reflects a different prediction taking exclusivity into account. If there are 2 red bars, the chance that the next bar will be green is 55.01%. Interestingly enough, this number stays in the high 50’s range for even seeing seven green or red bars in a row. Doesn’t look like flipping coins does it?… What would happen if we did this with coins?

coints

Well, as we expect, coins don’t have memory. The chance that the next flip will be heads after 5 heads in a row will still be roughly 50%.

In the hour data though, it seems to be that “stopping force” seems to be present. The chance is slightly higher than after a run up of 5 green bars, there will be a red to follow.

Of course, this study “supported” by opposing arguments. While buyers have to sell and sellers have to buy eventually, there is also momentum building that would push the next bars to be the same color. The chance that you have 5 green bars in a row in the Asian session vs London is quite different. These also have different meanings. There is no note of candle height. I believe this is why calculating waves will bring better variance in percentages. Hopefully my statistics are not simply done incorrectly.

Exploration of SB pt 1

I started with just playing with the data. In learning more about deltas and mapping price, I went ahead and took a ‘snap shot’ of what the SB(Signal Bender) waves show.

I calculated SB long to be H-PL(High-Previous bars Low), and SB short to be L-PH(Low-Previous bars High). Is this a way of sort of viewing “net price movement”? If the SB long line is higher than the SB short line, then by definition the difference between the high and the previous low is greater than the difference between the current low and the previous high. If there were a good way to visualize this, it would show the as long as SB long>SB short, there was some movement to the upside (at least in the high, not necessarily the close), disregarding inside bars. This concept took quite a bit of thinking, and it really depends on learning style. As a visual learner I used both my hands and pretended that the gap between my thumb and index were representing 2 adjacent bars. I imagined lines drawn from H-PL and L-PH, and thought about how the numbers changed as I changed bar height. Slightly Abstract. I find that this is a clever way to “blend” bars together. Keeping the essential parts (high/low) and ditching some minor details (time splitting bars, inside moves)

Anyhow, my sb waves seem to be different than the one Rel posted and I’m not sure how to resolve that. For now I’ll go on with my raw data and assume Rel is using additional methods in his basic sb indi.

Here’s what I got for 1 month of data (“snapshot”):

1 month

Edit: For fun here’s a comparison between the long/short TCDs and real price (via candle close) for the month:

fun

Like most data, it filters out into a ton of data points that fall into the average zone, a slightly smaller number of points that fall into the above average category, and a few that look like out liars, or points worth a second glance.

Extrapolating that to the larger data set, here’s what I found. I’m not sure what this could be useful for just yet:

data

SD: Standard Deviation

Also note that in the category of data points greater than 2x/3x dev, I’m only looking at the upper limits. I care about increase in activity, not decrease (atm).

Negative data points logically come from gaps. (ex. It means in the case of the SB long, that the High is less than the previous low. Even if the previous bar is bear and price closes at the low, the next open should be at the previous bars low already. If that bar moves straight down, the high-pl should still be at least 0; by being negative there must be a gap).

Interestingly the average SB Long and Short are the same; there isn’t a visible difference until you go out 6 decimal places. This has it’s own set of interesting observations, one being that price doesn’t seem to favor one direction over the other. Regardless of candle color, the movement is about 24 pips from H-PL. Memory of past statistics show that this is also the number from the difference of H-L (current bar!). This number is ALSO close to the average wick size on the daily chart. Hm…

I think the second point is kind of interesting. Average from H-PL: 24 pips. Average from H-L: 24 pips. Put this in a drawing. The next bar on average should be equal but opposite color from the bar before it?? (worth noting that the standard deviations between the two are different).

Movement Statistic

Movement

Pretty self explanatory:

How much movement is there in period “x” from the top to the bottom? In perfect execution, what’s the most you can expect to get away with in the respective time frames?

pip efficiency

Not surprisingly, pip efficiency goes down linearly as the time frames go by. This is explained intuitively by observing that higher time frames lend themselves to some form of retracement. 1m and 5m bars are most likely to complete without any wick.