Trading Strategy .edu



Evaluation of Earnings Based Trading Strategies: Surprises vs. Revisions

Presented to

Professor Campbell Harvey

February 24, 2000

Taco Bandito

Allen Born

Mark Davidson

Greg Dayko

Stephanie Shayne

Ed Shenkan

Executive Summary

Last year our group devised a trading strategy based on Earnings Surprises. While we achieved positive results, our group wanted to compare that strategy to a more forward-looking strategy based on quarterly earnings estimates. Thus, the purpose of this paper is to: (1) present the results of each trading strategy, (2) discuss in detail their implementation and design (including all assumptions), (3) evaluate their performance and potential weaknesses in the methodology, and (4) conclude which strategy is superior.

Overview

We examined three accounting-based earnings momentum trading strategies to determine whether an earnings surprise or an earnings revision strategy would generate greater abnormal returns. To this end, we evaluated earnings surprise and earnings revision data consisting of all stocks in the S&P 500 during the period March 1990 through June 1999. Our first strategy involved selecting stocks with very high positive or negative earnings surprises. Our next strategy involved looking at stocks with very large percentage increases in quarterly EPS estimates (i.e. from Q1 estimated to Q2 estimated EPS). Our last strategy involved selecting stocks based on what we called the “Taco Factor,” which is the sum of the earnings revision percentage in our second strategy and the percentage change in next quarter’s EPS estimate from the estimate for the same quarter the previous year (i.e. Q292 vs. Q291). For our methodologies, we constructed over 30 zero investment portfolios consisting of ten stocks per portfolio every three months (March, June, September, and December) for the time period mentioned above. We selected these months since most companies report on a calendar year basis; these months coincide with their quarterly reporting dates.

Our portfolios comprised five long and five (short) positions selected from those companies showing the greatest favorable (unfavorable) percentage earnings surprises and greatest favorable (unfavorable) percentage earnings revisions in a given quarter. We assumed a zero investment position in which we invested an equal amount of money in each asset. We generated the following returns. (See exhibits A & B for detailed portfolio returns).

Implementation and Design of Trading Strategy

Implementation

As mentioned briefly above, we employed an accounting-based earnings momentum trading strategy to earn abnormal returns. This strategy is accounting-based because we used publicly available accounting information to select the stocks in our portfolio. More specifically, we formulated a portfolio of equity securities using quarterly data from all of the companies in the S&P 500 from 1990 - 1999. Using IBES, we obtained information for:

• Mean quarterly earnings estimates

• Quarterly earnings surprises, based on mean analysts estimates

• Quarterly return data

The steps used in creating our portfolios were as follows:

1. Devised a list of companies that reported quarterly earnings surprises each quarter during the specified period. That is, actual earnings that differed markedly from the average of analysts’ expectations. For earnings revisions, we defined a revision as a change in estimate from one quarter to the next. Lastly, for the Taco Factor portfolios, we added to the revision above the percentage change in estimate for the upcoming quarter from the estimate for that same quarter the previous year.

2. Selected months in which companies who use a calendar year convention report quarterly earnings.

3. Divided the list of companies into “favorable” earnings surprises and “unfavorable” earnings surprises. In our earnings revisions methodology, examined the change in quarterly estimate from quarter to quarter and quarter to same quarter last year.

4. Computed the percentage difference between the actual earnings and the mean analyst forecast for surprises, calculated the percentage change in analysts’ estimates for our revision strategy, and calculated the percentage change in quarter estimate from same quarter last year.

5. Ranked the companies within each category by the magnitude of their percentage difference.

6. Selected five companies with the greatest favorable percentage difference and five companies with the greatest unfavorable percentage difference. In other words, we selected earnings surprises and revisions at the extremes of the respective distribution.

7. Formulated a zero investment portfolio of equity securities on the first day of the month following the earnings surprises and revisions by taking long (short) positions in stocks with favorable (unfavorable) percentage differences.

8. Held for six months (an intermediate trading rule time period) and then liquidated at the end of the holding period.

Zero Investment Portfolio

It is important to note that we constructed a zero investment (rather than a market neutral) portfolio to take advantage of the market mispricing. That is, we invested an equal dollar amount in long and short positions. However, we purposely did not construct a “market neutral” portfolio. A market neutral portfolio would have required that we balance the long position’s market sensitivity (beta) with the short position’s market sensitivity (beta). The advantage of such a portfolio is that it would have virtually eliminated market risk (except for the risk associated with the actual stock selection). However, the disadvantage of such an approach is that it would have necessarily changed either our equal weight component or our strategy from picking the top (bottom) five favorable (unfavorable) earnings surprises. As a result, we would not have benefited as greatly from the “extremes” in the earnings surprise distribution. Additionally, such an approach would require considerable more time (and judgement). That being said, the zero investment portfolios should yield slightly higher total returns due to the added beta (market) risk.

Assumptions

In designing and testing the three strategies, we made a few simplifying assumptions. First, we ignored trading costs associated with formulating our portfolio. Such costs would serve to reduce the total portfolio return. Second, we assumed adequate market depth (or liquidity) for our long (short) positions. We believe this is a reasonable assumption given the fact that our population consists of the S&P 500, which would suggest that these stocks are more liquid. Inadequate market depth would mean that taking long positions could drive prices up, while taking short positions could drive prices down. Third, we ignored the tax disadvantages associated with short-term capital gains. Tax implications are a critical component of investment strategy. Finally, when measuring the returns, we drew from a sample of data, which was neither size- nor market- adjusted. Based upon our data set, we felt that the companies in our pool were of adequate size to warrant a following by analysts. Therefore, although the companies may vary in relative size, we effectively rule out the “small size” effect.

Trading Rule and Methodology Observations

Evaluating Trading Results

When looking at our results, we see a number of interesting trends. First of all, the overall six month average return of 13.31% and 15.91% for revisions, and 37.12% for surprises were driven largely by performance across the sample of periods. This further translated into a 24% average return in excess of the average S&P annual return of 13% for earnings surprises. We also observed that Surprisers outperformed Revisions in twenty-eight of the thirty-eight months for simple revisions, and by twenty-eight of thirty-four months for the Taco Factor portfolios. There were several periods in 1991 as well as in 1998 in which we generated huge differentials between our surprise and revision portfolios. One possible explanation was that there were extraordinary events occurring at this time, which could have distorted analysts’ expectation. The Gulf War occurred in 1991, while the Russian Ruble and Brazilian Real collapsed in 1998. Analysts may not have understood the impact of these events on earnings. Finally, evidence suggests that the positive returns were driven more by the long positions than the short positions.

Drivers of the Returns

Positive v. Negative Returns: This strategy generated mixed results over the sample periods. Of the thrity-eight periods that we sampled, only five periods yielded negative results for earnings surprises, while the earnings revision strategy was success in twenty-nine of the thirty-eight periods. The average positive gain was 37.9% for surprises, 17.7% for revisions, and 22.9% for Taco Factor revisions, while the average loss was 3.2% for surprises, 16.9% for revisions and 11.2% for Taco Factor revisions.

Long or Short Positions: The overall return of the portfolio is definitively driven more by the long positions than by the short positions. The primary cause stems from the increased variability associated with the short positions. The short positions yielded negative returns in 47.4%, 44.7% and 61.8% of the surprise, revision and Taco portfolios, while the long position yielded negative returns in only 7.9%, 23.7%, and 17.6% of the surprise, revision and Taco portfolios. We believe this is attributable to the possibilities that companies may bury large expenses in one-time charges to create strong earnings for future periods. As a result, although they may report a poor one-time earnings figure, going forward performance will be favorable leading to growth in stock price. In addition, value stock strategists may see the initial share decline associated with an earnings surprise as an opportunity to procure value. As more shares are purchased, the price is driven up and the short position loses money.

Other factors which potentially drove the poor performance by our short positions include market forces and endowment effects. During the nineties, the Federal Reserve has maintained low interest rates, which has stimulated growth in the economy. This had the effect of increasing activity in the stock market and driving up stock prices. This stock price momentum may have caused our short positions to suffer (as their prices did not drop as precipitously as expected). Additionally, the shorts may have been impacted by a phenomenon known as the “endowment effect,” which states that people (and hence investors) respond more sharply to downside news than to positive news. This occurs because people generally dislike losses more than they like gains and believe that their wealth is more effected by small downside changes than by small upside changes.

The endowment effect may cause investors to respond more quickly to negative earnings surprises and revisions than to positive earnings surprises and revisions. As a result, by the time we formulated our portfolio (nearly a month later in some cases) we may have missed any trend downward in prices. In fact, the initial price decline following the negative earnings announcement may have been exacerbated by the endowment effect causing stock prices for these companies to trend slightly upward in subsequent months.

In the periods in which the portfolio generated significant positive return from both the long and the short positions, the average contribution from the long position was72% for both the surprise and revision strategies and 61% for the Taco Factor strategies. However, in the periods in which the portfolio generated negative returns, the negative driver was the short position in all cases for the surprise strategy, 55% of the cases in the revision strategy, and 57% of the cases in the Taco Factor strategy. This leads us to the conclusion that in general in a positive return scenario, the long position is the driver, but when the returns are negative or low, the cause stems from the short position.

Stragety and Methodology Weaknesses

We believe that implementation of the strategies presented in this paper will generate positive excess returns. However, we do recognize that potential weaknesses may exist in both the strategies themselves and in the method of testing and evaluating those strategies.

1. Revisions Not True Revisions: The major problem with our revision portfolios stems from the fact that the “revisions” we used are not true revisions. They are simply estimates for different quarters. As such, we are faced with such issues as timing of those estimates, seasonality, etc. The Taco Factor was an attempt to reduce those issues. Had we access to true annual EPS revision numbers our portfolios, and thus our findings, might have differed significantly.

2. Lack of Confidence in Return Numbers: We are not convinced that the return numbers obtained from IBES are accurate. The fact that we were able to obtain an average return in excess of 30% for the surprises portfolios certainly suggests that something might be amiss. Indeed, a random check of actual returns indicates that in some cases the IBES numbers are in error. Were we to continue this project, we would hand check the returns of each stock in each of our 110 portfolios (1,100 total returns), a process that could take several weeks.

3. Use of Current S&P 500 Stocks: The population of stocks that we worked with came from the current S&P 500. The issue here is that many of the stocks that are in the index now were not part of the index during the entire measurement period. Going forward, we would only be able to use the current S&P 500, thus a true historical sample would include looking only at members of the index during the measurement period. For example, our 1994 portfolios should come from the stocks that were members of the 1994 index. Unfortunately, we had neither the data nor the computing power to conduct this type of analysis. Simply using the current S&P index resulted in our analyzing over two million pieces of data.

4. No Scaling: When developing the strategies, we chose not to use the standard deviation of analyst earnings forecasts for purposes of scaling the earnings surprise. The primary reason for this decision was limited time and availability of data. However, we believe that, in the real world, using standard deviations of earnings estimates to scale the earnings surprise would likely produce a more robust trading rule than the ones presented in this paper.

5. No Provision for Extraordinary Earnings: In our strategies, we use actual reported earnings. We did not consider whether or not these actual earnings included any extraordinary gains or losses. As with the scaling issue previously discussed, the reason for this decision was lack of data. We do feel that incorporating this type of data into the trading rule might improve its predictive power. However, we feel that the impact of extraordinary gains on our trading rule would be minor because most companies that intentionally record extraordinary gains do so in order to meet analysts’ expectations. As a result, these companies’ earnings should be close to the analysts’ numbers. Therefore, we would not expect these companies to fall within the tails of our distribution.

6. Arbitrary Choice of Number of Holdings in the Portfolio: In the trading rule, we recommend buying the five stocks with the greatest (favorable or positive) percentage earnings surprise or revision and selling (shorting) the five stocks with the greatest (unfavorable or negative) percentage earnings surprise or revision. Although this choice did produce positive excess portfolio returns, we believe that an analysis involving either a larger or smaller number of long and short positions might produce even better results. For example, selecting the top 20% or bottom 20% of the distribution may yield greater abnormal returns. Obviously the optimum number of long (short) positions can be determined via modeling or calculus.

7. Arbitrary Choice of Holding Period: The trading strategy states that the long and short positions should be held for a period of six months. This choice was deliberate given the research performed by others on this method. Additional research would have to be performed to determine the degree of excess return generated in 1 day. Additional analysis would be required to test the degree of abnormal earnings based on a shorter reinvestment strategy.

8. Pre-Announcement Effect: In our trading strategy, we based our long and short positions on the magnitude of the percentage differences between actual EPS and average analyst forecasted EPS. However, it is common for companies to make pre-announcement warnings when earnings are expected to fall short of these expectations. When companies make pre-announcements, their stock prices will likely impound some of this information. For example, if a company pre-announced a negative expectation for its upcoming quarterly earnings, this would likely be absorbed in the price of the stock. As a result, the magnitude of any fall in future stock price may not be as great as it would have been without the pre-announcement. It is important to note that our trading rule would likely exclude such stocks during the formulation of our portfolio, as these stocks would not necessarily show up in the tails of the distribution.

9. Arbitrary Choice of What Months to Test Portfolios: Our original plan was to test portfolios formed in months that included the most quarter ends. (March, June, September, December). However, after viewing the data, we observed that many earnings announcements also occur during other months. While forming portfolios in other months does not detract from our model, it does increase the variability of returns as certain months may have relatively few observations when compared to others. A reduced number of earnings surprises would serve to decrease the number of outliers in the tails of the distribution and effectively reduce our abnormal returns.

10. Strong Assumptions: When testing the trading strategy we made several strong assumptions (no transaction costs, no liquidity constraints, and no tax implications). In practice, one would need to judge whether or not the excess returns generated by the trading strategy persist after relaxing these assumptions. For more discussion on our assumptions, see “Implementation and Design of Trading Strategy.”

Conclusion

Problems not withstanding, we believe that our earnings surprise and earnings revision strategies are effective if they can be implemented immediately after the announcement is made. If one was to choose between an earnings surprise and an earnings revision strategy (as defined in this paper), we clearly recommend choosing earnings surprises. Particularly for the short positions, timeliness is critical! Further, by utilizing a zero-investment portfolio, we constructed a strategy that is proficient in generating abnormal returns, while also minimizing some of the market (or beta) risk. Additionally, we found that our returns were more volatile for portfolios constructed around year-end surprise earnings announcements. This may be attributable to extraordinary charges included in those results as well as year-end “window dressing”. Therefore, we are confident that our strategy based on earnings surprises will generate positive alphas (and hence abnormal returns).

Exhibit A. Portfolio Results

-----------------------

[pic]

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

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

Google Online Preview   Download