# Re: Subject: Quantitative Analysis of High Entropy Economic Systems

From: John Conover <john@email.johncon.com>
Subject: Re: Subject: Quantitative Analysis of High Entropy Economic Systems
Date: 17 Feb 2002 18:53:17 -0000

```

The avg and rms do not change much-if the data set size used to
determine them accurately is sufficient, i.e., consistent with the
tsshannoneffective program. Tsinvest uses this methodology,
internally, and compensates the measured values of avg and rms for
data set size, prior to computation of P, which is what it uses as a
selection criteria for the stocks. That's why tsinvest did not chase
the dot-coms using the -d1 option, and why that option didn't lose
through 1999-2002. You can disable/enable the compensation with the
-c and -C options.

You can approximate the methodology for mental calculations. You know
that the chances of a "bubble" lasting t many days is erf (1 / sqrt
(t)), which is about 1 / sqrt (t) for t >> 1. So, as an approximation,
Pm = (rms + 1) / 2, and the effective P would be (1 - 1 / sqrt (1))
Pm; if P is less than 0.5, obviously, it would be a bad investment.

As an aside, note that an adaquate data set size for rms = 0.02,
(which is the median for all stocks on the US exchanges,) requires
about 10 years of daily data. Using a data set size smaller than that
is not investing-its gambling. But tsinvest will invest if the rms is
high enough to offset the losses it would take if it turned out to be
a bubble.

Tsinvest uses the exact same methodology that was used on GE. You can
have it dump the variables with the -r option, for all stocks in the
input file, so you can do it on k-mart, enron, etc.

John

Jeff Haferman writes:
>
> 1) The GE example... I wonder what the average daily increment (g)
> for GE would be, say, from 1/6/1962 through 2/15/2001 (i.e. 1 year
> less than what you present).  Then, what would the "gain curve"
> (the blue line) look like.  In other words, what sort of "predictive"
> capabilities might this sort of entropic analysis have? [eg, would
> data from 1962 through 2001 give a good idea of where we should be
> in 2002?]
>
> 2) I'd like to see this same sort of analysis applied to failed
> or failing companies (Sunbeam, K-Mart, Enron)...
>
> I'm not asking you to do this (unless you want to!)... if I
> find some time, I might do these experiments myself.
>
--

John Conover, john@email.johncon.com, http://www.johncon.com/

```