Different Lookback Period for New Highs and New Lows
Regarding 52-week New Highs and New Lows, what if you were to change the period for which new highs/new lows are calculated for the High/Low ratio? If you change it from 52 weeks into, say 1.5 times that, would you get a different reading?
It is an interesting question you pose. What you are talking about is a process known as “optimization”, which is often used in development of automatic trading systems. Say, for example, that a trader is trying to build a trading system around a 14-day RSI. He might try using a 13-day or a 12-day to see if he gets better results. He might also shift the entry and exit parameters (i.e. values of RSI for a trade signal) to see if he gets better results in terms of the profit/loss ratio, lower drawdowns, or some other measure of performance.
One problem with such optimization is that it can “tune” a system into a narrow set of conditions which only worked during one market period. That is why it is important to do “out of sample” testing once a system is optimized. This means testing the system on another time period of data which were not included in the sample set of data used for tuning the optimized parameters. If it does not work as well in the out of sample test, that is a sign that it is not a very robust system.
Your idea of changing the lookback period for calculating each stock’s status as making a new high or new low is interesting, but it is computationally impractical. It is an enormous task to maintain a database for calculating new highs and new lows, since one must keep track of every single stock on the exchange including all new IPOs, all splits and other distributions, all symbol changes, all delistings, etc. One must also track the existing 52-week high and low values for each stock, then run a program to test to see if today’s highs or lows exceeded those values, and then producing a report for each day’s values. If you wanted to try and optimize the 52-week lookback, then you would have to get the database to crawl backwards through mountains of data, and regenerate a new set of values for these indicators using the lookback period you have chosen.
Once you figured out the computing task, you would then be faced with the analytical task of figuring out whether the different lookback period results in data which are “better”. That would involve defining what “better” means, for your purposes, which is not as easy as it sounds.
It might be possible to model this with a smaller sampling of stocks, but even there the computing task is pretty big. Just to give you an idea of what I am talking about, I keep a spreadsheet file with the closing price and volume for each of the 30 DJIA stocks, going back to 1989. It calculates advances and declines, NH/NL, and Price Oscillators for each stock, plus a few derivative indicators based on those data. This single file is 42 MB and growing, and that is smaller than it might be due to the practice of converting formulas to values for past data to help save file space. If you consider that there are 100x that many stocks on the overall NYSE, and that you would have to keep track of OHLC data for each of them plus run formulas to calculate, it would be a huge task to undertake for any meaningful time period worth of data. It could not be done in a spreadsheet; one would need a very robust database program, and processing capability beyond that of even our modern PCs (or else a lot of time to wait for all that processing). You would need to run multiple permutations of the lookback periods you are interested in exploring, with analyses of the results of each.
So with all that in mind, it is usually a better course of action for us to do our best at learning to make use of the data that are available to us, rather than wishing for better data.
To learn more about using New Highs and New Lows, read these: