High Low Segment DataBase


Some Q&A's about the HiLoSegDB (High Low Segment DataBase) that others might be interested in.

quote:
You have stated the data was created from between 1.3 to 6.5 years of data. Presumably it was built up with data up to a certain date. What date is the probability tables drawn up to?


At the time that you asked this question the database had last been rebuilt about 1.5 years ago. Subsequent to your questions we ran a DB rebuild and the data used in each symbol now has a minimum of 3 years (for the YM) and a maximum of 8.3 years (for the ES) of data in the generation of the probabilities.

The rebuild was partly out of curiosity to compare the then and now and using a visual inspection there is very little change in the probabilities which is what we would expect. We would not expect this type of probability to change over the course of the market.

The build that was run today used data through to the end of trading on 4 August 2006.

quote:
My question relates to whether the probability tables data base will be
updated as the latest data become available.


No, generally not. As mentioned above there does not appear to be any point to this. There might be some curiosity in running this type of DB generation across certain time frames (for example run the DB for each calendar year) or across different time segment sizes. (example: Use 10 or 30 minute segments instead of 15 minute segments.) Each custom DataBase build is $20 with every 5th build being free (i.e. 5 builds for $80)

quote:
Has any buyer of the DB actually used it in their actual trading with any success and whether there are any statistics/reports from actual users.


We have not had any report backs from buyers with results. We have found that so far all buyers wish to keep their success (or lack thereof) private.

quote:
I realised that the price charged for the tables is not high but I know that if any trader depended on the prediction for probabilities, the trading result could prove many times more costly than the price of the product itself.


Correct. The tables/DB is not a trading system. It helps re-enforce a strategy with extra probabilities and they are also used for research purposes. To generate the software and DataBase took hundreds of hours so the cost of the DataBase is negligible compared to how much time it would take to create this type of information.

quote:
I would be very much obliged if you could give me some indication on the practical experience of the tables qualitatively or quantitatively. I am fully aware that a probability is not guaranteed to be a reality.


Well one way to quantify the usefulness of the tables is to do a forward test. In other words, you generate tables from data ranging from say 1/1/2000 to 12/31/2004. You then take those probability results and look at how accurate they are at predicting the probability of highs and lows during the year 2005. Are the percentage of highs and lows that happen in 2005 the same as suggested from the tables that resulted in the previous study.

We have not done that but based on the above results of rebuilding the DB we expect that this is what the data would reveal. One exercise which we have not done is the one suggested above in the custom builds where specific time periods are used. For example, isolating a start and end date which is in a bull market and also doing the same for a bear market.

I hope that this answered your questions.
quote:
Your comments on running the DB across different time frames sounds very interesting and I would be the first to support this idea. Have you considered a 5 minute segment DB as this appears to be a very common trading time frame for many traders or is this time too short to be of any real value...


Working out high/low probabilities could be done in up to 1 minute segments but anything less that 15 minutes is unusable and in fact I think that even 15 minutes might be too fine for this sort of thing and that 30 minutes might be more appropriate.

quote:
I do pay attention to the 81 minute chart in place of the more commonly used 60 minute of the ES. On this basis my preference for time segments will be [in minutes] 5, 10, 21 and 81 in addition to the standard 15 minute on which the DB has been built. Please let me know if these segments can be tailored for purchase.


I can set the generator to any value from 1 minute upwards - so the answer is yes to all of the above.

quote:
Your comments on testing in a bull and bear market are also extremely interesting.


The E-mini Data covers a good bull and bear market so picking 2 pairs of dates for either of those and generating a High/Low DB might be interesting. I'm guessing here but I would imagine that the bias would shift from beginning to end in each of these markets.

quote:
In respect of your comments to generate tables from data in a forward test are also most practical and are indeed the best way to satisfy oneself on the validity of the DB. There is only one MIGHTY BIG problem for myself. I don’t have any software that can do this and besides I don’t even know what software could do this kind of testing. I will be grateful if you could suggest a suitable software for this purpose.


There's nothing off-the-shelf that can do this - it would have to be custom made like the stuff that I have developed. The software is also tricky to run and that's why it doesn't get sold and I run tests for people. The results are generated as CSV files which I format in Excel spreadsheets so that further analysis can be done - for example using probability generated data to compare before and after a given date in the backtest / forward test scenario mentioned above.
30 August 2006

Hi Guy

Thank you very much for your respones to my queries.
You have been most helpful.

Best wishes and regards

Robert
You're very welcome.