Assignment 1 .edu



Assignment 1

Multifactor Predictions of Sector Returns

Global Asset Allocation

Team Alpha

Haley Lai

Noah Harris

Sanjun Chen

Nicolas Tollie

Introduction & Summary

We looked at different factors to see how well they can predict performance in different industries. Our goal was then to use the predictive power of these factors to establish scoring screens for each of the industries and test them out of sample to see if a good long-short strategy had been discovered.

We chose the basic industry groups manufacturing, services, financials, and utilities/transportation. The factors we chose were the next twelve months projected EPS growth, twelve months forward earnings to price ratio, current dividend yield, and inventory turnover.

While we were able to establish scoring screens for each industry for both equal weighted and value weighted portfolios, at best these screens provided mixed performance results out of sample and in some cases very poor performance.

Methodology

We selected our industries using the following SIC codes:

[pic]

As stated above, we used the following factors: twelve months predicted EPS growth, twelve months forward earnings to price ratio (E/P), current dividend yield, and current inventory turnover.

We limited our universe to stocks that trade on the NYSE and NASDAQ. We used 1/1990 to 12/1999 as our in sample period and used 1/2000 to 12/2004 to test our scoring screens out of sample.

We divided the industries up into 5 fractiles based on these factors and resorted the stocks each month to establish new portfolios. We then looked at the performance of these fractiles using the “heat maps” on Cam Harvey’s template. Our team members then subjectively assigned positive or negative scores (from -5 to 5) to the different fractiles based on good or bad performance. We took an average of the scores assigned by each of our teammates to get the final scores to use on our scoring screens.

We decided to establish separate scoring screens for equal weighted and value weighted portfolios, so each industry had two separate scoring screens. Finally, we tested these scoring screens out of sample to see how they performed.

Industries

The heat maps for the different factors are shown below with the scores shown above the fractiles that were scored.

Manufacturing

One year expected EPS growth for equal and value weighted portfolios:

As seen above, we scored fractile 4 a 3.5 for equal weighted and a 1.5 for value weighted.

Forward earnings to price ratio (E/P)

For this factor we only scored fractile 5 of the value weighted portfolio, but found it very significant and gave it a 4.5.

Dividend Yield

We scored several fractiles for both value and equal weighted portfolios. Looking at the dubious performance of equal weighted fractile 3 in 1997,1998 and 1999 made us question our data and prevented us from giving higher scores to this factor.

Inventory Turnover

We got strange results for this factor as it predicted very good returns for equal weighted fractile 5 and poor returns for value weighted fractile 5.

Services

One year expected EPS growth

The performance of these fractiles was pretty stable across both equal and value weighted.

Forward earnings to price ratio (E/P)

Results somewhat stable across equal and value weighted.

Dividend Yield

Equal weighted fractile 1 was not a great performer, but perhaps we rated it too low.

Inventory Turnover

Inventory turnover was not found to be a significant factor for service industry returns.

Finance

One year expected EPS growth

We only scored one fractile.

Forward earnings to price ratio (E/P)

Dividend Yield

Inventory Turnover

We scored two fractiles for value weighted. Perhaps we should not have considered this metric for financials as we are unsure of what a financials’ “inventory” is.

Utilities and Transportation

One year expected EPS growth

Not found to be a useful factor for this industry.

Forward earnings to price ratio (E/P)

Not found to be a useful factor for this industry.

Dividend Yield

We found dividend yield to be a very significant factor for equal weighted.

Inventory Turnover

We scored two fractiles for this factor.

Summary of Scores:

Out-of-Sample Performance of Scoring Screens

Heat map is shown followed by average returns, alphas, and betas of different fractiles.

Manufacturing

Our Equal weighted scoring screen performed well but our value weighted screen performed very poorly.

Services

Our equal weighted scoring screen generated low returns but the highest alpha. Once again our value weighted screen showed very poor performance.

Finance

Both of our screens performed very poorly for the financials. Perhaps none of our factors were that relevant to the industry.

Transportation/Utilities

Our value weighted scoring screen performed very well. It was the only fractile with positive performance.

Our equal weighted screen produced very strange results. Fractile 1 had by far the lowest returns, yet also had the highest alpha despite a comparable beta to the other fractiles. We are not sure what to make of this.

Conclusion

Our scoring screens produced mixed to very poor results out of sample. Often we did well with our equal weighted scoring screen but poorly with our value weighted scoring screen or vice versa. These results could be picking up some sort of size effect.

Given our mixed results, we cannot be sure whether our good and bad performers were the result of real sensitivities to our factors or simply by blind luck.

Possible Improvements

1. We should probably use a more conservative scoring system. First, we should probably only generate one scoring screen for both equal and value weighted portfolios. We should only score those factors that produce similar results across both equal and value weighted fractiles. Second, we should be more sensitive to the distribution of green and red cells on the heat maps. A middle fractile, even if it has a lot of green cells, may not be good if it is surrounded by red cells on both sides.

2. We could test more specific SIC code industry groups. Perhaps the groups we used were too broad to have real uniform sensitivities to different factors.

3. We might have selected different time periods for our in-sample and out-of-sample tests. 1990-1999 was mostly a huge boom time and 2000-2004 coincided with the bursting of the internet bubble and 9/11. The 1999/2000 cut off may have skewed our results.

4. Given more time, we should test more factors to make sure each industry will be sensitive to at least one of our tested factors.

5. We could use the optimizer to help us modify our subjective scoring system for the scoring screens.

While our results were not as we desired, we learned a lot as we went and feel we made a significant accomplishment in completing the process of quantitative stock selection.

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download