HAT'S THE BEST WAY TO SELECT STOCKS for your portfolio? You could always throw darts at the financial page, but that leaves a lot to chance. You might chart price movements and base decisions on recurring patterns, but that's not a whole lot better than the darts. Most folks would probably feel more comfortable basing investment decisions on the fundamental characteristics of the companies they're buying.

OK, now that that's settled, what fundamental factors should you use? There's historical data like Return on Equity or last years earnings. There's forward-looking projections like future long-term growth rates. Then there's ratios like P/E and Price to Book. Which ones should you use?

Obviously, you should use the ones that work best, the ones with the most predictive power. But which ones work best and do the same factors apply to all stocks? That's what we wondered and why we created our quantitative model.

As quants, we believe we can study what worked in the past to get a good idea of what will happen in the future. By finding what fundamental factors were correlated with previous price history, we hope to build a model that will help us select those stocks most likely to outperform in the future.

The Theory

Next, we narrowed down the fundamental factors to those we felt were important. If we were wrong in what we chose, our statistical analysis would let us know. If we didn't include factors that were important, we could always go back and add them later. The initial eight factors were:

• Return on Invested Capital -- Latest 4 quarters EPS / [(Average latest quarterly Long Term Debt + Stockholder’s Equity + 4 quarter ago Long Term Debt + 4 quarter ago Stockholders Equity)/Shares Outstanding]

• Pct. Long-Term Debt to Capital -- Long Term Debt / (Long Term Debt + Stockholders Equity)

• Price to Book Value -- Current Stock Price/ Worth of a company if it were liquidated

• Price to Cash Flow -- Current Stock Price/ ( Net Income + Depreciation)

• P/E Forward Four Quarters -- Current Price / Sum of the next 4 quarter’s earnings

• Earnings Long Term Future Growth Rate -- The Long Term Secular Growth Rate estimated for a period of five years

• P/E Long Term Future Growth Rate -- P/E based upon 2000 estimated EPS / Earnings LT Future Growth Rate. Must be a "prospective P/E" (i.e. not a 4 Qtr Trailing)

• Latest Earnings Surprise as a Pct. --
  1. Take the absolute value of the First Call consensus as the denominator.
  2. Take the difference, i.e. Report EPS - Consensus, as the numerator.
  3. Divide the numerator by the denominator. (the only times you will have "NM" is when the denominator = 0)*

The Procedure

We ran this procedure in two different ways. First using all the stocks of the S&P 500 and then again for each sector independently. As expected, the results differed in each instance.

Model Sector Weightings 6/30/00

Sector S&P 500 P3 P4 Basic Materials 1.9% 0.0% 2.2% Cap Goods 8.0% 3.2% 7.8% Communication Services 6.8% 0.0% 7.3% Consumer Cyclicals 7.4% 0.0% 7.4% Consumer Staples 10.3% 4.7% 9.4% Energy 5.4% 0.0% 5.2% Financials 12.7% 3.3% 12.6% Healthcare 11.6% 7.8% 10.8% Technology 32.8% 81.0% 30.8% Transports 0.6% 0.0% 2.8% Utilities 2.5% 0.0% 3.5% Source: Baseline

We then constructed two portfolios, one using the results from all 500 stocks (Portfolio 3), the other based on the top stocks from each sector (Portfolio 4). There was some overlap, but the sector based portfolio was much more balanced.

This truly is a work in progress. The eight fundamental factors and the resulting portfolios are just a first pass. As time goes on, we'll adjust the factors and maybe even the regression process to sharpen the models' effectiveness. If you have any suggestions, we'd love to hear them.

Portfolio 3

The resulting equation (shown above) has relatively strong explanatory power as evidenced by the the high R². (If you're not statistically inclined, an explanation of the meaning of R² is included in The Quantitative Approach.)

The only thing that gave us some concern was the high correlation between two of the five independent factors, price to book and price to cash flow (.7966). We'll need to monitor this in the future, but for now we let it go and proceeded to apply the equation to the S&P 500.

Our Quant Portfolios

Portfolio 3 Top 30 Stocks Based on Stepwise Regression Across All Stocks of the S&P 500 No Attempt is Made to Sector-Weight this Portfolio Rebalanced Every 60 Days Stocks Remain in the Portfolio Until Falling Below the Top 40 The Highest Rated Stocks Not Already in the Portfolio are Added When Existing Constituents are Removed Portfolio 4 Top Stocks of Each Sector Based on Stepwise Regression of Each Individual Sector of the S&P 500 Number of Stocks Selected in Each Sector Determined by Current Sector-Weightings of the S&P 500 Rebalanced Every June Stocks Remain in the Portfolio for 12 Months Unless Deleted for Special Circumstance e.g. Acquisition Stocks Removed for Mergers and Acquisitions are Replaced by the Next Highest Rated Stocks in Their Specific Sector

We got our factor values from Baseline, plugged them in, and calculated expected return values for each stock. We then sorted the results from the highest to the lowest. Now we were ready to construct our portfolio.

To do so, we simply took the top thirty stocks. Why thirty? No reason, other than it's a workable number to track and has the potential to provide enough diversification to eliminate non-systematic risk.

Over 80% of the stocks in this portfolio were from the technology sector. In fact, only 5 of the S&P's 11 sectors were represented (see table,, above). This really shouldn't come as a surprise since the model is designed to pick stocks with the best 12 month performance potential. Very few would argue that consumer cyclicals or utilities will be great short-term performers.

Going forward, we'll adjust this portfolio (P3) every other month around the 15th of the month. Stocks with rankings that have fallen below 40 will be removed and replaced with those not already in it with the highest rankings. Although there won't ever be more than 30 stocks in the portfolio, we'll let their rank fall below 40 before removing them to hold down turnover. Since P3 will be rebalanced six times a year, it should be more dynamic than P4.

Portfolio 4

With this in mind, we went through the same procedure for each sector as we initially did for the entire S&P 500. As you would expect, we got different regression equations for each sector. The accompanying table shows which factors were significant for each as well as the coefficient values.

We constructed P4 by selecting the top-rated stocks in each sector. We didn't impose a limit on the number of stocks, but rather kept choosing until the sector weightings were roughly equivalent to those of the S&P 500 as of June 30, 2000. We ended up with 52 issues. (Current holdings in both P3 and P4 are available at these links.)

This portfolio won't be as dynamic as P3. Since it's more sector neutral, we intend to let it run until next June without major modifications. Given we're attempting to determine its predictive power for one-year returns, it only makes sense to let it have a year before passing judgment. The only anticipated changes over the next twelve months will be additions or deletions due to mergers and acquisitions.

Observations

It's logical to assume there would be differences in the factors affecting various sectors. This was borne out in P4. You'd also think the overall S&P would be affected differently than any individual sector. This was also confirmed given the differences in the two portfolios.

P3's excessive technology weighting was also anticipated given that growth parameters were favored over value. Technology is where the growth has been and where the model predicts it will remain for the next 12 months. The sectors not represented in the portfolio have traditionally been slow-growth areas.

But surprisingly, growth factors were not overly emphasized in P4, the sector model. The PEG ratio and earnings surprise -- two favorites of growth investors -- only fell in the middle of the pack. But then value factors such as return on capital, price to book, and forward P/E didn't do too well, either.

The biggest surprise was the fact that plain old long-term future growth rate was the clear winner, appearing in 7 of the 11 sectors. No other factor came closer than 5.

Does this mean growth is quantitatively more significant than value? Maybe it does, or maybe it just means growth has outperformed value over the past decade when the data for the models were compiled. Over the next several months, we'll get a better picture as we monitor and tweak the model.

In the meantime, we'd like your help. What would you like to include or change in the model? Yes, you too can become part of this work in progress. E-mail your comments.

E-mail your comments.

Search this site! Just enter you key word or words:

 

Get current quotes or follow your own custom portfolio, courtesy of E-Line Financials:
 

Search:TickerName Company DataQuickQuote

 

Homepage

Return to Top