Saturday, February 21, 2009

Ponzo's Time Machine

A while back, I posted a "linkfest" from Euan Sinclair's excellent book, "Volatility Trading." One of the links was to an eclectic financial time-series analysis and mathematics website maintained by Prof. Peter Ponzo at (yes, you read that correctly).

The good Professor's site is quite literally packed with thought provoking material, including the focus of this post: a search engine that seeks to match the past 25-week's price behaviour to the most similar period back over fifty-years of historical data according to the maximum pearson coefficient.

Investors worldwide are currently grasping to estimate price support for the major equity indices. The problem they face, is that the indices have come so far down that one needs to look back more than ten-years to apply traditional support-level analyses.

Now, this post isn't about the merits or demerits of technical analysis, but given this problem, I thought it would be entertaining - at the very least - to crank up Ponzo's matching engine to find the current period's most analogous historic epochs. To accomplish that, I slightly modified Ponzo's code to output the five most relevant epochs, ranked from most to least mathematically analogous. The highest ranked historic six-month periods turn out to have commenced on (Yahoo! Finance Charts allow date entry):

  1. May 5, 1969
  2. September 14, 1987
  3. August 21, 2000
  4. May 6, 2002
  5. May 13, 2002
So -- really four periods here excluding the May 2002 overlap. Nevertheless, if that isn't entertaining enough for you, Ponzo's Time Machine goes one step further, projecting the current time-series forward by four months based on the the matched epoch's subsequent market behavior.

The marked dark blue line in the chart below represents the most recent six-month period for the S&P500 Cash Index, while the marked red line represents a four-month forward projection using the top five matches from above weighted according to rank, as follows:

(Chart should read February 2009)

As shown, all but one projection series ended lower four months down the road, the interesting exception being the 1987 match. In fact, the weighted average projection suggests a further -5% decline from last Friday's (Feb 20) close by the end of the outlook period. The aggregate range indicates a projected S&P500 low of 660 (-14%) and a high of 800 (+4%). Not exactly thrilling prospects, and I can only imagine what the results would have looked like had I extended the analysis back through the 1930s!

Granted, this current economic crisis is like no other in terms of its root causes and rapidity of associated market declines. At the same time, we are currently highly oversold and well overdue for a powerful bounce, but Ponzo's indicated matches do provide us with an interesting historical range to ponder in an otherwise vacuous data point arena where both technical and fundamental analysis fail us. If there is enough interest in this study, I will update it periodically as we move through this difficult time.

o More on Nearest Neighbor Analysis, submitted by TraderFeed.

Never Investment Advice


Peter said...

I happened across your "Ponzo Time Machine" article and found the idea of projecting the "best five" absolutely delightful!!

Now why didn't I think of that!?

Peter Ponzo

jgpietsch said...

Hi Peter, good to hear from you again. Thanks for stopping by. Best, Jeff