Long-Short Strategy for US Small Cap Equity Trading



Long-Short Strategy for US Small Cap Equity Trading

Alex Volzhin

Kader Hidra

Dan Grundman

Damian Olesnycky

Jason Trujillo

Discussion of Objectives and Results

Goals

Our goal was to utilize analytical tools (Fact Set) to develop a monthly trading strategy for US small cap equity that will generate significant alpha but also keep risk at level not far above that of the broader US Equity market. We decided to examine the small cap universe for a couple of reasons. First, we feel as though less is known about the factors that influence small cap equity movements than large and mid cap stocks. This disparity allows us to garner a greater advantage through the use of Fact Set. Second, the greater volatility of small cap stocks allows for greater returns using an effective long/short strategy. Lastly, we looked at this exercise from the perspective of a small fund. Generally speaking, smaller funds have an advantage in this universe especially with regard to active trading strategies because they do not need to take as large a position in each of the portfolio companies. As such, we considered this universe to, perhaps, be less efficient and thus a place where we can capitalize the most on the tools we have available.

Method

The Models

Our intention was to use the Fact Set Alpha Test 3 to test our hypotheses on the predictive value of different factors on returns in our selected universe. All factors chosen were first tested as a single variable initially. Those that demonstrated a strong directional trend were then matched up with other variables and bi-variate models were run in the same. We then went further and ran several multi-variate models. Finally, we took our most successful models and ran a subjectively scored alpha test. All models were tested in our in-sample date range described below. Successful models were then run in an out sample date range. The results for our different models are given in the results section of this report.

General Screening Criteria

Our universe for stock selection meets the following criteria:

Security Selection – Our analysis is strictly limited to US common stock.

Market Capitalization – We included companies that were between $300Mil and $2Bil in size. We kept the range fixed through time. However, we feel that if one were to expand on this initial model it would better to have the range adjust over time to account for inflation.

Volume – All companies considered have an average daily volume greater than or equal to 500K shares per day. Given that our strategy will require monthly rebalancing, it was decided that sufficient liquidity is a necessity.

Benchmark – S&P 500. In deciding how what to use as our benchmark many different indices were considered. Given that we are looking at small cap stocks certainly an index like the Russell 2000 would be a good choice. However, we decided that we wanted to compare our performance to a more widely used stock index. This allows the average investor to more easily appreciate the performance of our strategy.

Historical Periods Used in Analysis

For our analysis we used in an in-sample period of ten years to test the effectiveness of our predictive factors and then ran a one-year out of sample test.

In Sample – The in sample data consists of a ten year period from the beginning of January 1995 to the end of December 2004.

Out of Sample – The out of sample test was run in a one year period from the beginning of January 2005 to the end of December 2005.

Factors Tested

We tested a wide range of range of factors related to fundamental, momentum, economic and technical analysis. Of the factors we tested we found only the momentum factors were of any predictive value. Given more time there are a number of other factors that we would like to test, in addition it would be of benefit to further verify our results.

Successful Factors

1 Month Return – This was our most successful factor and we found it to be quite effective as both a positive and negative indicator. As a monthly trading factor it was lagged one month and compared to the subsequent month’s performance. This factor was chosen because it thought it would be a good way to capture the momentum behind stock price movement. The thinking goes that in many cases with regard to small caps a good month will be followed by another good month and likewise a poor month will be followed by another poor month.

6 Month Return – This factor was also very effective, however it was effective as the 1 month return. In addition, it was highly correlated with our 1 month factor and as such was not used in any of our scoring models or in our out of sample tests. As with the 1 month return we lagged this factor by one month as well. We choose as a longer-term measure of momentum. Again, we figured that on the balance, positive movements would predict future positive movements and do the same for negative trends.

Current Price to

52 Week High – This factor also proved to be quite effective. It was lagged 30 days. We choose this as a way to give a perspective on the current price level relative to past performance. We especially that this factor worked well when combined with either of our return factors.

Unsuccessful Factors

Book to Price – We were hoping that this fundamental value factor would provide some indication of future prices moves but after running a model with this factor we found no persistent trend. That is, performance was relatively similar among the fractiles.

P/E – With price to earnings we did not expect it to be a very strong factor with regard to small cap prices. However, we thought it would be prudent to test it with a model. As expected, we the results were not meaningful.

Revision Ration – With this factor we compared the number of upward revisions to downward revisions. Our expectation was that we would be able to predict short-term price movements from this ratio. Unfortunately, this was not the case. We believe that the amount of coverage among this asset class is a possible reason why this factor did work well.

Unemployment – As well as looking at fundamental, momentum and technical factors we also wanted to examine economic indicators that may be able to provide some insight. We expected that unemployment would be a good indicator as to the strength of the economy, with a strong economy being a positive indicator for small cap prices. After running our model through our in-sample universe we did find that unemployment was a significant indicator.

Several other factors were taken into consideration but we did not feel the findings warranted discussion. In many cases we found that the data for past years was simply not available for some factors that we wanted to test. In other cases we were simply trying variations of the factors given above.

Results

One-Factor Models

Our best one-factor models delivered good returns (monthly arithmetic average):

• 1-Month Returns Model +6.98%

• 6-Month Returns Model +4.26%

• Price to 52-Week High +3.55%

1-Month Returns

[pic]

6-Month Returns

[pic]

Price to 52-Week High

[pic]

Unsuccessful Factors

As we mentioned above, we have tried several other factors, but they proved to be useless for forecasting. On the following slides, there is no clear trend among the fractiles’ returns.

Book to Market Price

[pic]

Price to Earnings

[pic]

Revision Ratio

[pic]

Two-Factor Models

Although one-factor models looked promising, they all suffer from inconsistent performance. During certain months, losses are big enough to drive the manager out of business. Therefore, we tried to improve the models by additional screening factor. The performance of the models is summarized below:

• 1-Month Return & Price to 52-Week High +6.95%

• 6-Month Return & Price to 52-Week High +4.55%

Bivariate Model: 1-Month Return & Price to 52-Week High

[pic]

Beta for Bivarate P to 52High & 1 Month Return Model

[pic]

Bivariate Model: 6-Month Return & Price to 52-Week High

[pic]

Multivariate Model

[pic]

Scoring Screen

We chose our two best factors to run a scoring screen: 1-month return and the Price to 52-week high.

We subjectively assigned scores between -5 and 5 for each factor and for fractiles 1 and 5 in order to infer a long-short strategy. We will go long on stocks in fractile1 and short on stocks in fractile 5.

For 1-month return:

Fractile 1: +5

This fractile has the highest average return.

48% of the months have the highest returns.

5% of the months have the lowest returns.

Fractile 5: -5

This fractile has the lowest average return.

50% of the months have the lowest returns.

11% of the months have the highest returns.

Price to 52-week high:

Fractile 1: +3

This fractile has the highest average return.

28% of the months have the highest returns.

22% of the months have the lowest returns.

Fractile 5: -3

This fractile has the highest average return.

37% of the months have the lowest returns.

23% of the months have the highest returns.

In-Sample Two-Factor Model: 1-Month Return & Price to 52-Week High with Scoring

[pic]

Out-of-Sample Two-Factor Model: 1-Month Return & Price to 52-Week High w/o Scoring

[pic]

Out-of-Sample Two-Factor Model Beta: 1-Month Return & Price to 52-Week High without Scoring

[pic]

Out-of-Sample Two-Factor Model: 1-Month Return & Price to 52-Week High with Scoring

[pic]

Out-of-Sample Two-Factor Scoring Model Beta: 1-Month Return & P to 52-W High

[pic]

Long/Short Distributions Positively Skewed After Scoring

[pic]

BI-VARIATE Heat MAPS out-of-sample

|P52 & 1-Month Without Scoring Out-Of-Sample Quintiles |

|Fractile |1 |2 |3 |4 |5 |

|Summary |1.17 |-0.13 |-0.18 |-0.16 |-1.95 |

|12/31/2004 |-2.36 |-6.16 |-10.88 |-9.26 |-16.48 |

|2/01/2005 |7.44 |-2.72 |1.86 |0.51 |-2.83 |

|3/01/2005 |-4.90 |-3.94 |-3.06 |-3.68 |-7.37 |

|4/01/2005 |-4.66 |-8.84 |-8.23 |-7.71 |-9.81 |

|4/29/2005 |6.54 |7.95 |6.44 |7.96 |16.91 |

|6/01/2005 |2.59 |1.82 |2.08 |3.78 |1.27 |

|7/01/2005 |6.53 |8.48 |8.06 |8.89 |7.10 |

|8/01/2005 |0.15 |-0.59 |0.69 |-0.19 |-5.82 |

|9/01/2005 |1.31 |1.60 |-2.10 |-2.37 |0.27 |

|9/30/2005 |-4.20 |-6.38 |-3.99 |-5.29 |-9.54 |

|11/01/2005 |6.13 |8.23 |6.42 |7.12 |9.80 |

|12/01/2005 |0.70 |0.87 |2.50 |0.38 |-2.12 |

|Geometric Mean |15% |-2% |-2% |-2% |-21% |

• Fractile 1 has the highest average return.

• Only 3/12 months have the highest returns.

• Here we are concerned by these 2 months where we actually got the lowest returns in fractile 1.

• Fractile 5 has the lowest average return.

• 8/12 months have the lowest returns.

• Here we are concerned by these 2 months where we actually got the highest returns in fractile 5.

• The Long/Short spread is satisfactory: 36%

|P52 & 1-Month With Scoring Out-Of-Sample Quintiles |

|Fractile |1 |2 |3 |4 |5 |

|Summary |5.59 |0.95 |0.39 |-2.27 |-6.48 |

|12/31/2004 |-3.42 |-4.40 |-7.35 |-16.83 |-15.23 |

|2/01/2005 |4.98 |5.91 |1.93 |-1.73 |-7.00 |

|3/01/2005 |1.21 |-4.00 |-3.57 |-10.67 |-8.57 |

|4/01/2005 |-0.76 |-2.39 |-8.23 |-13.13 |-15.52 |

|4/29/2005 |11.01 |7.14 |7.29 |17.00 |6.75 |

|6/01/2005 |5.47 |2.29 |3.41 |-2.37 |-0.48 |

|7/01/2005 |14.22 |6.17 |8.27 |6.34 |4.05 |

|8/01/2005 |6.10 |-1.87 |-0.20 |0.13 |-7.36 |

|9/01/2005 |6.40 |2.19 |0.11 |0.41 |-10.91 |

|9/30/2005 |0.51 |-3.04 |-4.34 |-7.27 |-16.36 |

|11/01/2005 |18.57 |5.21 |7.32 |6.43 |0.11 |

|12/01/2005 |4.88 |-0.80 |1.70 |-0.75 |-3.80 |

|Geometric Mean |92% |12% |5% |-24% |-55% |

The scoring screen alleviates our concerns:

• Fractile 1 has the highest average return and outperform the unscored screen by far!

• Fractile 1 has the highest average return. 10/12 months have the highest returns.

• Fractile 5 has the lowest average return and underperformed the unscored screen by far!

• Fractile 5 has the lowest average return. 9/12 months have the lowest returns.

• The Long/Short spread is satisfactory: 147%.

BI-VARIATE Heat MAPS in-sample

P52 & 1-Month Without Scoring In-Sample Analysis:

|P52 & 1-Month Without Scoring In-Sample Quintiles |

|Year |1 |2 |3 |4 |5 |

|1995 |56% |85% |70% |40% |-16% |

|1996 |60% |59% |5% |12% |23% |

|1997 |24% |25% |45% |10% |27% |

|1998 |46% |25% |48% |9% |-5% |

|1999 |150% |229% |76% |91% |103% |

|2000 |52% |123% |6% |-37% |-75% |

|2001 |3% |45% |4% |48% |-47% |

|2002 |-14% |-33% |-53% |-50% |-43% |

|2003 |60% |93% |82% |72% |148% |

|2004 |23% |16% |-3% |3% |-3% |

|Arithm. Mean |46% |67% |28% |20% |11% |

• Fractile 1 has NOT the highest average return.

• Only 3/10 years have the highest returns.

• Here we are concerned by 2003 when we actually got the lowest returns in fractile 1.

• The spread would have crushed us!

• Fractile 5 has the lowest average return.

• 5/10 years have the lowest returns.

• Here we are concerned by 2003 when we actually got the highest returns in fractile 5.

P52 & 1-Month With Scoring In-Sample Analysis:

|P52 & 1-Month With Scoring In-Sample Quintiles |

|Year |1 |2 |3 |4 |5 |

|1995 |108% |38% |48% |16% |11% |

|1996 |31% |27% |26% |36% |16% |

|1997 |77% |39% |30% |32% |-25% |

|1998 |57% |23% |37% |8% |-1% |

|1999 |131% |126% |112% |128% |104% |

|2000 |138% |17% |41% |-69% |-75% |

|2001 |69% |9% |37% |-40% |-47% |

|2002 |9% |27% |-41% |-47% |-75% |

|2003 |176% |94% |86% |76% |26% |

|2004 |72% |21% |5% |-12% |-33% |

|Arithm. Mean |87% |42% |38% |13% |-10% |

• Fractile 1 has the highest average return.

• 8/10 years have the highest returns.

• The scoring eliminates the 2003 crush!

• Fractile 5 has the lowest average return.

• 10/10 years have the lowest returns.

Possible further analysis:

Value VS Equal weighted analysis.

Concerns

Transaction Costs

Given that this is an active trading strategy, transaction costs are of great concern. The strategy calls for monthly rebalancing. After looking at several examples from our report we found that in many cases you would see substantial turnover of the securities in both the 1st and 5th fractiles. In addition, to regular turnover the number of securities held in the portfolio, while not overly large, is still significant (~60 per fractile). This is of considerable concern because we cannot simply use a futures contract to change our expose given that this is a specific portfolio.

Clearly, transaction costs could be a concern for our strategy. One possible way to mitigate this problem would be to consider longer holding periods. Specifically, one may want to consider the efficacy of the a 3 month and 6 month holding period while still employing the same strategy.

Short Selling Constraints

Considering that we are dealing with small cap stocks, some of which are very thinly traded short selling the proper amount of a given security may at times become a challenge. We setting our screen we tried to mitigate this problem by setting the minimum allowable volume to 500K shares daily. However, even at the level we still believe this to be an issue. At the very least, short selling the necessary may require working with several dealers and could have cost implications.

Execution

As is the case with all theoretical strategies, executing it properly in practice may pose problems. Our model assumes that we will be able to take receive the data on the last trading day of the month and then complete all of our trades at the closing price on the first day of the month. In practice this will certainly pose a challenge. As such there are certain issues with trading near holidays or on Fridays that may alter our results.

Volatility and Exit Signals

Generally speaking small cap stocks are very volatile. As we examined the in sample, there were several months in which our strategy would cause substantial losses. As a fund manager such losses may not be tolerable. In addition, as our current model is structured to be run monthly we do not have any interim exit strategies in place. Considering that we are taking on considerable exposure to common stocks an unexpected shock wave could have devastating results. Further, we do not have plan in place for liquidating the portfolio if the market faces any sort systematic movement that may be adverse to our position. Before implementing this proposal as an active strategy it would be good to have such protocol in place.

Fact Set

As we consider the value of our analysis it must be noted that we are using a tool with which we have had limited experience and which is extremely complicated. Before moving forward with the strategy we would first like to audit our data and rerun our models to be sure of their accuracy. In addition, we are generally placing a good deal of confidence in the quality of our data. Considering that it is taken from multiple sources this component may be in question. Specifically, there may be problems with the data timing between when Compustat logs a particular variable and say when I/B/E/S does. Given our overall limited knowledge we would also want to test this out before moving forward.

Limitations

Investment Scale

Given the type of securities that we are examining and the number of securities in our proposed portfolio it is questionable whether or not this strategy could be implemented by a large fund. Certainly, any investment over 500 million dollars would not be effective. Every time the portfolio was rebalanced the prices of many of the securities would be moved significantly especially with the regard to those on the lower end of capitalization range.

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

[pic]

[pic]

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

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

Google Online Preview   Download