Sunday, May 12, 2013

Predictive Analytics

I am going to cut right to the chase.

I had hoped to write something funny about how I don't like predictions, but I don't think anybody actually wants to read that. And I'm not good at writing that kind of stuff anyway.

All I really want to talk about is the table you see below.

In almost all of my trading I use actual past trades or back-testing (or a combination of the two) to make decisions. It's all about probabilities.

But what information should I use to make trading decisions about new stocks that have only recently gone public?

That is the problem I'm trying to solve here.

So the approach/experiment I've chosen is to model the weekly price behavior of 10 stocks that have gone public over the last year and a half (beginning with their IPOs) to see if I can make reasonably accurate predictions about how the stocks will perform after we get a buy signal.   

The table includes the 10 stocks, the dates and prices of their initial buy signals, and their predicted average weekly returns. It also includes the most recent closing prices for all 10 stocks, which we use to compute the actual weekly average returns, actual total returns, and Compound Annual Growth Rates.

Company (Symbol)Buy Signal DateBuy Signal PriceClosing Price 5/10/2013Predicted Average Weekly Return %Actual Average Weekly Return %Actual Total Return %CAGR %
3D Systems (DDD)1/30/2012$14.09 $43.85 +0.63%+2.07%+211.2%+144.6%
Marathon Petroleum (MPC)6/25/2012$44.13 $77.38 +0.39%+1.32%+75.3%+91.4%
Phillips 66 (PSX)8/13/2012$42.33 $61.38 +1.15%+1.08%+45.0%+66.3%
Facebook (FB)11/26/2012$28.00 $26.68 -0.00%-0.08%-4.7%-10.3%
Caesars Entertainment (CZR)12/3/2012$7.32 $15.02 -0.90%+4.95%+105.2%+446.8%
LinkedIn (LNKD)1/7/2013$118.00 $173.78 +0.61%+2.49%+47.3%+226.8%
Pandora (P)1/7/2013$11.02 $15.58 +0.06%+2.24%+41.4%+188.4%
Zillow (Z)1/22/2013$35.35 $54.14 +0.74%+3.22%+53.2%+338.3%
Zynga (ZNGA)2/4/2013$3.43 $3.23 -1.16%-0.35%-5.8%-21.4%
Splunk (SPLK)4/8/2013$42.26 $44.89 +0.66%+1.58%+6.2%+119.2%

So what is the table telling us?

A lot. This blog would go on forever if I chose to break down everything it's telling us.

So I'm going to cut to the chase again and look at just 4 simple solutions.

1. The base case. In the interest of full (and embarrassing) disclosure, the first solution is what I've actually done. I took the Facebook, LinkedIn, and Zynga trades, but none of the others. Why those stocks? I honestly don't know. Gut feeling. And my gut feeling is obviously worthless. Heck, I knew that 10 years ago. That's why I use a system to trade. In any case, the result of this approach is an average gain of +12.2%.

2. Take all of the trades. No real decision to make here. The average gain with this approach is +57.4%, so it's already a much better alternative to what I've been doing.

3. Take only the trades with a positive predicted average weekly return. Now we're getting somewhere. This would include Phillips 66, Zillow, Splunk, 3D Systems, LinkedIn, Marathon Petroleum, and Pandora. Our average gain here is +68.5%.

4. But wait, there's more. What if we took only the trades with a positive predicted weekly alpha? This would eliminate Pandora, because it's predicted average weekly return of +0.06% lags the predicted average weekly return of the S&P 500, which is +0.22%. This last approach gives us an average gain of +73.0% across 6 stocks.

Clearly there is some room for improvement regarding my current methodology...

No comments:

Post a Comment