EDIT: 7/20/2014: The trading tools and dashboards are no longer part of the site.
This is a tool I developed only a few weeks ago so I’ll be testing it live here using the FOREX market. The Weekly Projection tool applies statistics to the 100 most recent weekly bars to project the price movement for the upcoming week. The projection is in the form of a graphic showing the projected high, low and close of the coming week along with the order in which the high and low will occur. This will be available on the Free Dashboard for the EUR/USD and in the Member’s Dashboard for other selected pairs.
Below be dragons…and math (gasp)!
The general hypothesis is that one week’s price action (as described by our dependent variables) is affected in part by the prior week’s price action (as described by our independent variables).
I developed two nominal level independent variables which describe the price action of each bar. These are the OC classification and the HL classification systems. Let’s look at these in detail.
The OC system classifies the bar’s price action entirely in terms of where the open and close are on the bar (open/close, so it’s OC, see? Si, I see!). Calling the low of the bar 0% and the high 100%, I divide the bar into three pieces. Anything above 70% is the high end, anything below 30% is the low end, and the 40% in the middle of the bar is…well, it’s the middle of the bar. So a bar that opens in the middle portion and closes in the low end would be coded “ML” under this scheme. There are 9 possible classifications; HH, HM, HL, MH, MM, ML, LH, LM, and LL.
Note that if I had retained the actual percentage location of the open and close, I would be working with ratio level data here instead of nominal level data. So I’ve given up some information content in the data in exchange for a system that’s easier to use. If you’re unfamiliar with the distinctions between different levels of data, take a look at my post on Levels of Measurement.
The HL classification system is based on the locations of the bar’s high and low in relation to the prior price bar, and also on the order of the high and low events. For example, a given week’s price action might consist of a higher (than last week’s) high followed by a higher low, coded as “HHHL.” Or it may consist of a higher low followed by a lower high (an inside bar) coded as “HLLH.” Note that in order to see whether the bar’s high or low came first, I have to zoom in on the shorter period bars, in effect looking “inside” the weekly bar. There are 8 possible classifications under this system; HHHL, HHLL, LHHL, LHLL, HLHH, HLLH, LLHH, and LLLH.
The OC and HL classification systems are logically independent of each other, meaning that a bar with any given OC pattern could also have any one of the HL patterns (Try it! There are 72 possible pairings). However, this does not mean that they are statistically independent of each other. Each OC pattern may be more likely to be paired with certain HL patterns and not others. I haven’t studied this in detail yet.
Once I classified each bar in terms of the independent variables, I used
four dependent variables to classify the price action of the following bar.
The first variable is just whether the high occurs below the low or not. This is a binary (yes or no) nominal level variable.
The other three values are the locations of the high, low and close of the bar expressed as differences from the previous bar’s close divided by the previous bar’s range. That was a mouthful so here’s an example.
Bar A: High = 40, Low = 30, Close = 32
Bar B: High = 35, Low = 28, Close = 31
We want to calculate our three variables for bar B, using the previous bar A as a reference scale. Bar A’s range is 10.
High = (High – Prev. Close)/Prev. Range = (35-32)/10 = 30%
Low = (Low – Prev. Close)/Prev. Range = (28-32)/10 = -40%
Close = (Close – Prev. Close)/Prev. Range = (31-32)/10 = -10%
Another way to state this is:
The high of bar B is 30% of bar A’s range above bar A’s close.
The low of bar B is 40% of bar A’s range below bar A’s close.
The close of bar B is 10% of bar A’s range below bar A’s close.
I think we could argue that these are ratio level values, but it won’t matter. We’ll only be taking the medians and means of these values which is fine for anything at interval level or above.
I built an Excel spreadsheet to do all the calculations described here, so I just have to update the price data each week. The price data consists of the open, high, low and close of the most recent 100 weekly bars along with a manually entered HL pattern code. As I mentioned above, I have to get that by zooming “inside” the weekly bar to see its price action.
The week just completed will have an OC classification and an HL classification. We want to project the next bar’s price movement based on what has happened after similar bars in the past.
The spreadsheet first finds all bars in the data with the same OC
classification and looks at the price action of the bar following each one. From those, it finds the medians of the high, low, and close dependent variables. It also counts the number of times that the high occurs before the low and vice-versa for those bars. It then does the same thing for all the bars with the same HL classification.
To determine the final high, low and close values, it takes the mean of the two medians found for each classification scheme. To project the order of the high and low, it just takes the greater sum of the two.
Got all that? Ok, here’s a real life example from the weekend during which I’m writing this post.
For the week that ended on Friday 3/14 (Pi day!!! gotta buy apple pi..), the EUR/USD opened and closed in the middle part of its range, so the OC pattern is MM. Relative to the prior week it had a higher low followed by a higher high, so the HL pattern is HLHH.
Looking at bars with the OC pattern first, there were 5 with the MM pattern in the data prior to this week. The median “close to high” move was 77% of the previous bar’s range, median “close to close” was 40% and median “close to low” was -56%. We only actually have 3 MM bars with the next bar’s price action zoomed in the data so far, and of those, the high occured first 2 times and the low occurred first just once.
Looking at the same data for the 13 prior HLHH pattern bars we have a median “close to high” move of 59%, “close to close” of 19% and “close to low” of -24%. The high occurred before the low 5 times and the low occurred first 8 times.
So first of all the “low before high” configuration wins by 9 to 7. Given the previous bar’s high, low and close of 1.3967, 1.3834 and 1.3911, we project the open, high, low and close of the next bar as follows:
Open: Just the prior close, which was 1.3911
(77% + 59%)/2 = 68%
1.3911 + 0.68(1.3967 – 1.3834) = 1.4001
(-56% + -24%)/2 = -40%
1.3911 – 0.4((1.3967 – 1.3834) = 1.3858
(40% + 19%)/2 = 29.5%
1.3911 + 0.295((1.3967 – 1.3834) = 1.3950.
And voila! That’s where the projection comes from. How well will these projections aid us in our trading? We’ll find out. The EUR/USD projection for each week will be in the Free Dashboard. Under this Tools/Weekly Projection sub-category, I’ll also be posting graphics of the actual price performance plotted against the projection each week so we can see how well (or poorly) they work.