How to get rid of gaps in Price ROC

[jamesmarkof]

In chapter 3 of "Value in Time", there is a section that talks about "Price Gap Corrections". It makes a good point regarding price gaps when calculating price rate of change. It states that price gaps throw off the divergence analysis between ROC and ER. Here is a quote from that section:
"Figure 3.14 shows for Darden Restaurants the price rate of change that was represented in Figure 3.5, but for which price gaps have now been eliminated."

My question is how exactly did you eliminate price gaps? How did you smooth them out? The chapter does not talk about any specifics.

Thank you.

[Pascal]

Ji James,



Thank you for asking. This is a rather technical issue that I indeed avoided to detail in the book.
It is however very simple. This "gaps" issue was important in order to calculate the Price/LER divergence. The problem was that a large 10% gap for example on a company earnings would push the divergence indicator to an extreme.

Since I work on minutes data, I take a rolling window of for example 2000 minutes, which allows me to rate the strength of a price ROC and the LER strength. This means that I have a price ROC indexed variable of 2000 values as well as a LER(2000) index variable for LER.

In order to work on the price ROC by eliminating the gaps, I only need to build a new price ROC variable that calculates all the gaps in the last 2000 minutes and subtract the gap level from the Price of the minutes. A new ROC can then easily be calculated.

For example: if in the past 2000 Minutes, there were 5 price gaps of for example +1$, +1.2$, -2$, +0.5$ and -0.3 $, then the total of these gaps is +0.4$. If the actual price at minute 2000 is for example $100 and the price at minute 1 was $90, the ROC is $10/$90. After taking the gaps into consideration, the new ROC will be $9.6/$90. I hence have effectively eliminated the 0.4$ cumulative gaps during that 2000 minutes period. I do that for every minute.



Pascal

PS: By the way, I seldom use the divergence analysis anymore in my daily trading. It is not as good as the Active Boundaries/LER combination or the Supply/LER combination.

[jamesmarkof]

Makes sense. Thank you.

When you say "I seldom use the divergence analysis anymore in my daily trading", do you mean you seldom use it when you day trade? In other words, do you still use it for swing trading? Or do you just mean in general the divergence analysis does not work as well any more?

[Pascal]

Makes sense. Thank you.

When you say "I seldom use the divergence analysis anymore in my daily trading", do you mean you seldom use it when you day trade? In other words, do you still use it for swing trading? Or do you just mean in general the divergence analysis does not work as well any more?

In the past few years, I have been only using Active Boundaries/LER combination or the Supply/LER combination.

One disadvatnage of the divergence analysis is that one needs to tune the divergence to the volatility difference between price and volume. This is a difficult process that is indeed non-reliable and should be manually checked. Hence, when I scan a large number of stocks, this divergence analysis could point to stocks that are in fact not interesting. LER, AB and Supply are easier to tune. I tune them ones a week.



Pascal

[jamesmarkof]

Good point about the volatility difference.

I guess the reason I found AB a bit sketchy is because of the active float value. One has to discover it for each and every security that's being monitored. I know you generalize it by saying that it can just be the amount of shares exchanged in the past 60 to 90 days, but I found, as did you, that it's not that straightforward. The best way is to simply match reversals in AB to reversals in price, and THAT is tedious process. Unless, of course, there is another tool for that.

[Pascal]

Good point about the volatility difference.

I guess the reason I found AB a bit sketchy is because of the active float value. One has to discover it for each and every security that's being monitored. I know you generalize it by saying that it can just be the amount of shares exchanged in the past 60 to 90 days, but I found, as did you, that it's not that straightforward. The best way is to simply match reversals in AB to reversals in price, and THAT is tedious process. Unless, of course, there is another tool for that.

Yes, it is tedious, but the Active Float is calculated only once in a few months and the calculation is computerized.


Pascal

[jamesmarkof]

But how do you computerize the calculation of the Active Float value? Meaning the number of shares. Don't you have to put in a few values and then compare of the resulting active float curve to the actual price curve within the same time frame to see if reversals mostly match in both graphs? How do you automate this process?

[jagin]

But how do you computerize the calculation of the Active Float value? Meaning the number of shares. Don't you have to put in a few values and then compare of the resulting active float curve to the actual price curve within the same time frame to see if reversals mostly match in both graphs? How do you automate this process?

+1 for the question.

Pascal would you mind to put some light on this?

Regards

Jarek

[Pascal]

But how do you computerize the calculation of the Active Float value? Meaning the number of shares. Don't you have to put in a few values and then compare of the resulting active float curve to the actual price curve within the same time frame to see if reversals mostly match in both graphs? How do you automate this process?

Exactly! This is not an easy process, but it can be computerized.

I have two maintenance programs running for the AB system:

1. One a year, I reset the active float.

Usually, I take as active float an amount of shares that has been exchanged between 60 days and 90 days.
- If the float turnover is lower than 60 days (for very actively traded stocks) then I need to take the total float as the active float
- If the float turnover is much higher than 90 days (sluggish/defensive companies), I use max 90 days, but still check that the AB pattern follows the price pattern.
- For other stocks, then I need a trial and error that tests different floats. (It is a pattern recognition software)

2. One per week, I check that a new boundary is not setting (new trend starting)


Pascal