Overview - Fuqua School of Business



SHARPE MINDS ASSET MANAGEMENT

RAYMOND JACOBSON

VENKATA KODALI

JOSH ROSE

JIM SHEEHAN

JOHN WATTS

GLOBAL ASSET ALLOCATION

THE FUQUA SCOOL OF BUSINESS, DUKE UNIVERSITY

ASSIGNMENT 1

SPRING 2001

Overview

For our analysis, we decided to investigate an optimized momentum/contrarian strategy. The "basic" strategy involves going long a momentum portfolio and going short a contrarian portfolio. However, we attempt to optimize this simple (and frankly, ineffective) strategy by developing a predictive regression model, which provides a signal for optimal switching.

Methodology

We gathered monthly return data on 27 S&P sectors, making sure we picked a diverse sample of sectors. We then filtered the returns to find the top five and worst five sectors, with respect to monthly performance. Finally, momentum and contrarian portfolios were constructed. For the momentum portfolio, we picked the five sectors at time (t) that performed the best during

time (t-1). The contrarian portfolio was just the opposite, i.e. the five sectors at time (t) that performed the worst at time (t-1). A combined portfolio was created from going long the momentum portfolio and going short the contrarian portfolio. Since we are dealing with diverse sectors, which in turn are made up of a variety of individual securities, we make the assumption that the combined long/short portfolio is market neutral, i.e. portfolio beta equals zero.

Data

Data was collected for the time period from April 1, 1980 through February 1, 2001. For consistency, we used the first day of the month for all of the data, except when as average value was used as was the case with bond and bill data. In collecting data for these dates, we made sure to evaluate the appropriate monthly data.

For example, returns for May 1, 2000 are really returns for April 2000 (April 1 through May 1). Of course, we recognize that the first of the month is not always a valid trading day, but it provided us with a consistent data set. Almost all of the data was obtained via DataStream. The other source was the Federal Reserve web site, which provided monthly averages for all of the bond data (both government and corporate). The following S&P500 sectors were used.

|S&PTRUC(PI) |S&PBREW(PI) |

|S&PEPOW(PI) |S&PSOFT(PI) |

|S&PBANK(PI) |S&PENTE(PI) |

|S&PINSM(PI) |S&PTOFD(PI) |

|S&PCHEM(PI) |S&PRETF(PI) |

|S&PAUTO(PI) |S&PTOBA(PI) |

|S&PPAPR(PI) |S&PMACW(PI) |

|S&PAERO(PI) |S&PMACW(PI) |

|S&PELEC(PI) |S&POILD(PI) |

|S&PTELE(PI) |S&PDRUG(PI) |

|S&PTOYS(PI) |S&PHOSP(PI) |

|S&PHOTL(PI) |S&POFFC(PI) |

|S&PPUBL(PI) |S&PCOND(PI) |

|S&PRETS(PI) | |

Model

In order to apply a predictive aspect to this strategy, we attempted to predict when the Long Momentum minus Short Contrarian strategy was effective, as well as when it wasn't. When the strategy produced a positive return, we assigned a "1" to that result. A negative return was assigned a "0." This yielded a set of 0's and 1's on which we ran a logistic regression, in the hopes that we could predict the best strategy for a given time. The Regression Output section presents our model with further discussion.

Variables

We considered a wide range of variables shown in the table below, from those affecting the economy as a whole to those more focused on the stock market and a particular style. Ultimately, the only statistically significant variables were the Changes in the US Gov't Yield Spread (between the 30yr and 5yr), and the Changes in the Corporate Credit Spread (Moody's Baa - Aaa).

The table below shows a list of variables considered for regression analysis; we also considered variations on these, such as net and % change. The first two items (shaded) are the ones we used in our final model. Their significance is discussed in more detail in the Regression Output section. In our evaluation, all variables were lagged by one period, except for the economic indicators, which were lagged by two periods. Our model achieved greater than 53% accuracy over the entire period. Out of sample analysis produce the same results (53%).

|Variables |Economic Significance |

|Moody’s Baa – Aaa Corporate Credit Spread|During good times, more people invest in lower grade bonds, resulting in lower yields and|

| |narrower spreads. A narrower credit spread likely favors growth stocks and momentum, |

| |whereas a widening spread ?? |

|US Govt 30Y – 5Y Bond Spread | |

|US Govt 5Y - 3M Bond Spread |Same as above but focuses on the short end of the curve, which may shift more rapidly |

| |than the long end. |

|S&P Composite Dividend Yield |We thought that the Dividend Yield for the Composite might have some predictive power, |

| |yet it did not result in any statistically significant correlation. |

|S&P Composite PE Ratio |Likewise, we thought that the PE Ratio for the Composite might have some predictive |

| |power, yet it did not result in any statistically significant correlation. |

|S&P Growth to Value Book to Price Ratio |This was even more of a surprise, as we examined the Book to Price ratio of Growth vs. |

| |Value for the Barra Indices. Book to Price gives an indication of portfolio Again, no |

| |statistically significant correlation. |

|Consumer Price Index | |

|US Unemployment | |

|US Housing Starts | |

|US Consumer Confidence | |

Switching Strategy

Once we had our model, we could then implement an enhanced trading strategy. The output from the logistic regression corresponded to probabilities of the LM-SC portfolio producing a 1 (i.e. positive return) or a 0 (negative return). We decided to go Long Momentum/Short Contrarian for values greater than 0.6. For values lower than 0.4, we were Short Momentum/Long Contrarian.

If the value fell in between this range, we invested in T-bills for that month. The resulting strategy generated monthly returns of 0.82% (10.34% annually) over the 20-year horizon. In comparison, T-bills only produced monthly returns of 0.54% over the same horizon. As mentioned above, our strategy results in a market neutral portfolio; hence, the appropriate benchmark becomes the risk-free rate as measured by T-bills.

[pic]

[pic]

Regression Output

Lagged Change in Moody's Baa - Aaa Corporate Credit Spread: Our regression shows a negative coefficient for this variable, which is in agreement with out intuition. If the change is positive (the spread is increasing), it will result in a lower output. The lower output favors "0," the contrarian strategy, which makes sense ince a widening spread may imply contrarian stocks are oversold.

Lagged Change in the US Govt 30Y - 5Y Bond Spread: Our model also specifies a negative coefficient for this yield curve indicator. Similarly, as the long-short gap increases, we expect the contrarian approach to be favored since it may be a sign that a recovery is ahead.

Conclusions

Our results show an average annual return of 10.34% for the optimized strategy. The average annual return for investing in T-Bills is 6.65%. Our strategy did not take transaction costs into account. If transaction costs are taken into consideration it may be better to just invest in T-Bills which have a lower volatility.

Surprisingly financial variables (book/market, P/E, Dividend Yield) failed to have predictive power. On the other hand credit spreads did well, possibly signaling when riskier contrarian stocks would outperform.

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

[pic]

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

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

Google Online Preview   Download