From: John Conover <firstname.lastname@example.org>
Subject: script that does the same thing as tsinvest?
Date: 14 Dec 2000 18:07:53 -0000
The numerical methods used in tsinvest are quite straight forward, and most can be done in a spread sheet. There is, also, a set of 60 some programs that can be used for general fractal analysis at http://www.johncon.com/ndustrix/utilities.html, or they can be downloaded as a tape archive from http://www.johncon.com/ndustrix/archive/fractal.tar.gz. The tsinvest sources were, largely, cut-and-stick from the sources to these programs. A possible scenario to automatically pick stocks might be: To find the marginal increments, (i.e., the fluctuations,) of a stock's price, use the tsfraction program. To find the average increase in value, use the tsavg program. To find the risk of the investing in a stock, (i.e., the root mean square of the fluctuations,) use the tsrms program. which is the two values used by tsinvest. So, if sprice is the name of a file containing the price of a stock over time: tsfraction sprice | tsavg -p would print the average increase in value of the stock, avg, and: tsfraction sprice | tsrms -p would print the risk of investing in the stock, rms. Then, the Shannon Entropy, (or Shannon Probability, i.e., the likelihood of an up movement,) P, would be: P = ((avg / rms) + 1) / 2 Or you could use the tsshannonwindow program to measure the entropy directly: tsshannonwindow -a -b -c -d -e -f -g -h sprice To calculate the gain, G, in value of a stock, use the tsgain program: tsgain -a avg -r rms which, as a stock picker, bigger values are better. All it does is solve the equation: G = ((1 + rms)^P) * ((1 - rms)^(1 - P)) This would be the same as the -d1 option to tsinvest. The -d2 option assumes avg = rms * rms, and the -d3, rms = sqrt (avg). To get the -d5 value, one needs the persistence in the time series, and the tsrootmean program can be used: tsrootmean -p sprice will give the persistence, (which is the same as the Shannon Entropy, P, in this case.) If you want to compensate the values for uncertainty do to data set size, use the statistical estimation program, tsstatest: tsstatest -P P -D 0.001 which is the default in tsinvest, (the -c option disables it.) Or, if compensation for run lengths is required, (the -C option in tsinvest,) use the tsshannoneffective program, (where count is the number of time samples in the sprice file,): tsshannoneffective avg rms count Both the tsstatest and tsshannoneffective programs print a probability of uncertainty do to data set size, so the measured Shannon Entropy and one, or both, of these values can be multiplied together to get the likelihood of an up movement in a stock's price, which is compensated for data set size and/or run length. If you automate this with a shell script, and do it for many stocks at the same time, (picking stocks that have the largest value from tsgain,) the result will be very close to what tsinvest does. John BTW, note that there are really only two empirical constants used, avg, and rms, both of which can be calculated in a spread sheet from a stock price time series. Many spread sheets have statistical estimation capability, and the functionality of the tsshannoneffective program can be approximated with the square root function, for t >> 1. -- John Conover, email@example.com, http://www.johncon.com/