Project Summary - Fuqua School of Business



PROJECT SUMMARY

Objective

Our objective for this project is to apply what we have learned in the classroom to practice, and get a hands-on experience on quantitative stock selection process using FactSet. By exploring various factors and scoring method, we hope to identify long and short portfolios that could consistently generate positive return with minimal market exposure.

Method

We focused on sorting method in our stock selection practice. Our screen universe is 500 largest companies in US, as we believe that large companies tend to have more extensive and reliable financial data. Our rebalance period is one month. Our in sample period is from Mar. 95 to Dec. 2002, and our out of sample period is from Jan. 2003 to Dec. 2005.

Ten factors were screened. Those factors are price momentum, expected earning yield FY1, expected earning yield FY2, FCF yield, dividend yield, short position, price to book value, change in consensus, CFO growth, and unexpected earning. Four factors were selected based on various diagnostics. Due to time constraint, we only considered allocating assets into quintiles. We then performed both subjective and objective scoring system to sort companies based on total score.

In our objective scoring, we used solver to optimize scoring weights for various fractiles that we selected. Furthermore, we created dynamic weights by introducing synthetic assets by multiply dummy variable Term Structure (0 for inversion and 1 for no inversion) and asset return.

Results

Four factors, expected earning yield FY1, expected earning yield FY2, FCF yield, and price momentum, consistently outperformed other factors in our in-sample period from 1995 to 2002. Those four factors were chosen for further analysis.

We have explored two scoring methods in our final model. In our subjective scoring method, weights were assigned to various fractiles, and objective weights were solved using optimization. We also created synthetic assets by multiply dummy variable, term structure, with return. The analysis indicated that the dynamic weights outperformed our subjective scoring, judged by our diagnostics and heat map.

Future Direction

If time permits, we would like to further our study in the following area.

• Trading cost: Our analysis did not take trading cost into consideration. But to develop a trading strategy in real world, we would have to incorporate that into the model.

• Migration track: In our scoring system, we did not track the movement of various equities between fractiles. If implemented appropriately, migration track would improve our current scoring method. Due to time constraint, we did not have time to implement it in this project.

FACTORS

Expected Earnings Yield (FY1)

Formula:

IH_Meadian_FY1R(0)/CM_P(0)

Pros and cons for choosing the factor:

This forward looking E/P ratio incorporates a firm’s expected earnings in the next fiscal year and its current stock price. The ratio allows us to screen stocks based on a firm’s future profitability normalized by its price, which is consistent with the forward-looking characteristics of stock valuation. On the flip side, accounting earnings may not as closely reflect the intrinsic value of the firm as do other measures such as free cash flow.

Results:

In our in-sample period, the expected earnings yield factor has generated quite impressive results: with a long/short strategy, we generate positive returns in 6 out of 8 sampling years. Not surprisingly, in the bubble years of 1998 and 1999, our strategy of longing 1st fractile stocks and shorting the last fractile stocks gives us negative returns. On average, the factor works well in that its first fractile average performance is the best while that of the last fractile is the worst.

|1995 | |127.8 |129.4 |126.0 |

Ave. performance:

Average returns:

[pic]

Alpha:

[pic]

Expected Earnings Yield (FY2)

Formula:

IH_Meadian_FY2R(0)/CM_P(0)

Pros and cons for choosing the factor:

FY2 is quite similar to FY1 in that they both incorporate earnings expectations and they are both subject to the same limitations of using accounting earnings. However, we feel that FY2 is more forward looking and therefore, might add more value to the stock screening process.

Results:

Except for the bubble years (98 and 99), the factor generates even better results than does FY1 in that the difference between the average performance of the 1st fractile and that of the 5th fractile is bigger, and in 1996, the 1st fractile becomes the best performing one.

|1995 | |128.5 |130.0 |122.4 |

Ave. performance:

Average returns:

[pic]

Alpha

[pic]

Price Momentum

Formula:

CM_RET(-1 -13)

Pros and cons for choosing the factor:

We choose to test this factor because over the past, some stock returns seem to follow certain patterns. Using price momentum may allow us to incorporate past price movement trends into our predictions of returns. This factor, however, does not incorporate any future expectations of a stock’s performance, and therefore, may raise concerns over its capability of being used as a predictive factor.

Results:

Price momentum consistently segregates top fractiles as the better-performing ones from bottom fractiles as the worse-performing ones in all but one year (2000). We reason that this is because we used past price moment in our stock screening, and in the bubble year of 1999, this strategy is damaging our portfolio performance given the backward-looking characteristic of the factor. Nevertheless, this factor generated impressive results with a long/short strategy.

|1995 | |130.2 |125.4 |123.6 |

Ave. performance:

Average returns:

[pic]

Alpha:

[pic]

FCF Yield

Formula:

CQ_CF_FREE_PS(0L45D)/CM_P(0)

Pros and cons for choosing the factor:

Free cash flow might be a better measure for a firm’s earnings than net income in that non-cash transactions are removed from earnings calculations. We lagged the FCF for 45 days to account for the lagged release of financial data.

Results:

This factor generates quite impressive results over the in-sample period. According to the heat map, the first fractile is the best-performing one overall, while the last fractile is the worst.

|1995 | |133.8 |128.9 |126.1 |

Ave. performance:

Average Returns:

[pic]

Alpha:

[pic]

Other Factors

In addition to the above four factors, we have also tested six other factors, including: short position percentage, dividend yield, price to book value, CFO growth rate, change in consensus, and unexpected earnings. Since these eight factors did not generate as consistent results as the above four, we decided to move on with establishing the scoring system with the abovementioned four factors.

MULTIFACTOR SCORING ANALYSIS

Scoring Factors Selection

We choose four best performing factors based on the outcome of individual factor analysis. These four factors include a fundamental factor: Free cash flow yield (FCF Yield), expectational factors: Expected Earning Yield (FY1), Expected Earning Yield (FY2), and a momentum factor: Price Momentum (one year lag return).

Judgmental Scores

Price Momentum:

• Value weighted performs better than equal weighted

• Fractile 1: +3 : Highest average return with consistent performance from 1995 to 2002, except in 2000

• Fractile 5: -3 : Lowest average return with consistent performance from 1995 to 2002, except in 2000

FCF Yield:

• Performs well in both value weighted and equal weighted measurement

• Fractile 1: +4: Highest average return with consistent performance from 1995 to 2002. Incurs smallest loss in 1999 compared to the loss using other factors.

• Fractile 5: - 4: Lowest average return with consistent performance from 1995 to 2002. Realizes lowest gain in 1999 compared to the loss using other factors

Expected Earning Yield (FY1)

• Equal weighted performs slightly better than value weighted, especially in 1999 and 2000

• Fractile 1: + 2 Highest average return with relatively consistent performance from 1995 to 2000. Incurs highest loss in 1998 and 1999

• Fractile 2: - 2 Lowest average return with relatively consistent performance from 1995 to 2000. Realizes highest return in 1998 and 1999

Expected Earning Yield (FY2)

• Equal weighted performs better than value weighted, especially in 1999 and 2000

• Fractile 1: + 2 Highest average return with relatively consistent performance from 1995 to 2000. Incurs highest loss in 1998 and 1999

• Fractile 2: - 2 Lowest average return with relatively consistent performance from 1995 to 2000. Realizes highest return in 1998 and 1999

Scoring Screen

The long/short strategy provides a consistent result under scoring screen, i.e. only a 9.4% loss in 1999 at in-sample test. We notice that this strategy incurs a loss of 18.9% in 2003 at the out-of-sample test. In general, this portfolio has a 13.7% annualized average return and a 15.0% standard deviation from 1995 to 2005. The sharp ratio is 0.92 and the alpha is 19.78%. The averaged return for S&P500 is 20.3% with a 16.1% standard deviation. The Sharp ratio is 1.26. Please see “table scoring 1” for the heat map

[pic]

(Table: Scoring 1)

Summary

Scoring screen provides a consistent strategy after taking into account of importance factors. The outcome of the subjective scoring screen depends on the experience of the portfolio manager. We also notice that long fractile 2 and short fractile 5 would potentially generate a larger return under current scores. Please refer to Scoring Graph 1 and Scoring Graph 2 for details on annualized average return and Alpha.

[pic]

(Scoring Graph 1)

[pic]

(Scoring Graph 2)

DYNAMIC WEIGHTS ANALYSIS

Methodology

Based on our factor analysis and heat map characteristic

• Choose seven different quintiles. Free cash flow yield quintile 1, 5 ( FCF_1, FCF_5), Two year expected yield quintile 1,5 ( FY2_1, FY2_5), One year expected yield quintile 5 (FY1_5), Price momentum quintile 1,5 ( MM1, MM5).

• Choose dummy variable. Add a dummy variable of one month lagged term structure ( 3 month T Bills – long term Treasury Bond). When yield curve inverses, the dummy variable equals zero, other wise equals one.

• Run the optimizer with the above 14 “assets”. Multiple the chosen seven assets by the dummy variable to get seven synthetic assets accordingly. Use the solver to find the 14 weights.

• Dynamic weight setting. If the yield curve inverse, we adopt the “synthetic assets” weights; if not, we use the sum of two weights. All the above weights are based on the original value weighted testing results.

• Run scoring with dynamic weight. Add the term structure to the screen model, and assign the different weights as scores to the scoring model, and re-run the scoring.

Results

The dynamic testing result is listed as following.

[pic]

Compared to the original factor models, which had significantly underperformance during the IT bubble years, this dynamic weighted model could pick up the market opportunity during 1999 and 2000. Also, this model could also mitigate the risk during bubble years. The average ranking difference between quintile 1 and quintile 5 is as high as 2.6. For out–of- sample test from 2003 to 2005, quintile one is the best performing fractile .

In this objective scoring method, the performance is significantly improved compared to the subjective scoring method. The long short portfolio generated by this method has a return of 12.0%, compared with 13.2% of benchmark. The sharp ratio of long short portfolio is 1.52, higher than 0.87 of benchmark.

The following is the performance summary of different models:

[pic]

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

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

Google Online Preview   Download