Global Asset Allocation and Stock Selection



Global Asset Allocation and Stock Selection

Assignment 1

Stock Selection Model

Bright Sun Asset Management

Nigel Anderson

Mattias Lundahl

Jakob Midander

So Sugiyama

February 2006

METHODOLOGY

Our universe screen limits the stocks to US equities with a lower limit on market capitalization of $2.5 billion. This gave a sample of 875 companies. By using such a market cap limit, the resulting sample is basically an extended version of the S&P500. Therefore, we are using the S&P500 index as the benchmark for our portfolio.

The time period used is from January 2000 until December 2005, thus avoiding the internet bubble in the late 1990s.

Our model uses the factors described below and divides the companies sampled into monthly performance quintiles based on these quintiles. The total data set is then downloaded and processed, creating summary output for the quintile portfolios to be used for evaluating the different factors. The equally weighted and market cap weighted returns are calculated for each quintile portfolio in each month. These two sets of return series are then analyzed for patterns of comparative performance of the individual quintile portfolios.

Based on this analysis, subjective scores are assigned for belonging to a specific quintile for a specific factor. The scores are summed for each company and each month and the total score is then used to create new quintile portfolios.

Our investment strategy is then to go long the top quintile and short the bottom quintile according to the total score ranking. The performance of this strategy is shown below.

OUT-OF-SAMPLE TESTING

Due to limited data processability and the voluntary constraint to start from January 2000, we have not performed out-of-sample testing of our strategy, but used almost the entire data range from 2000 up until today as sample for building the model. The sample is then used to determine reasonable scores for belonging to particular quintiles for the respective factors we are using.

To test the model out of sample would mean using the same scoring system and applying it to data from other years along the following lines:

1. Gather data starting January 2006 (or later) for the same screen (US equities with market capitalization > $2.5 billion) and the same factors as before.

2. Assign scores to each stock at each time depending on the quintile it belongs to for each of the factors. The scoring pattern is the same as the one described above.

3. Sum the scores for each stock in each month and create quintiles for each month.

4. Apply the strategy previously used - i.e. going long the top quintile and going short the bottom quintile.

5. Assess the performance of this long-short strategy.

FACTORS

As discussed above we used three factors in our analysis. There follows an explanation of these factors, the reasons why we included them in our model and some of the potential limitations these factors have which need to be considered when we use the model.

Earnings Yield

Earnings yield is effectively the same measure of value as the omnipresent P/E ratio. It allows us to compare companies on the basis of their historical earnings and their current share price. In essence, it tells us whether one company is more expensive than another.

It is a superior measure to dividend yield which does not take into account retained earnings and which is subject to managerial bias. For example, companies will only make changes to dividends if they can be sustained in future as the market looks unfavorably on companies who chop and change dividends. Our initial analysis used dividend yield, but the large number of growth companies declaring no dividends makes comparison impossible.

In theory at least, earnings is an accounting measure which is audited and routinely reported. However, there are a number of issues which should be considered when using such a measure:

- earnings can still be manipulated by management

- while it can compare value companies, it does not necessarily pick up growth companies whose earnings are significant lower due to high uncapitalised expenditure

- comparisons across sectors can also be problematic

- it is backward looking; a more relevant, but uncertain variable, may be to use forecasted earnings

Our analysis of the earnings yield over the period provides some interesting results. In an equally weighted portfolio, using the historic earnings yield solely as a predictor of future returns gives some strong results with the long portfolio outperforming the short portfolio in five of the six sampled years.

As we would expect, this factor is weaker when a portfolio is constructed based on market capitalizations. Intuitively, this makes sense because our predictor is looking for companies who are under market capitalized. Going overweight in those we consider overvalued makes little logic.

Price to Book

We chose the price to book ratio (P/B) as a means to identify ‘value’ stocks. The book value, defined as a company’s net assets minus its outstanding debt, gives a snapshot of the value per share if a company shuts down its operations, sell of its assets, and pay its debt. A low P/B ratio implies that the stock price is low relative the value of the company, and, hence, an attractive stock to buy.

There are many reasons to be cautious when using the P/B-ratio as an input for stock selection. In some industries (e.g. IT, financial firms) tangible assets are relatively low, resulting in high P/B-ratios. In these cases, relying on the P/B-ratio solely could lead to missed investment opportunities. Also, a company’s profitability depends on the usefulness of its assets; a company may have invested heavily in assets that are of little use for running its business successfully. For these reasons, it makes sense to study the development in the ratio for a specific stock over time. For example, a stock that is trading at an all-time-high P/B level could imply that the stock is overvalued. Additionally, investing based on P/B ratios might be less successful in bull markets, when investors appreciate (risky) growth oriented stocks, whereas ‘value’ stocks tend to do well more bearish conditions.

While we appreciate the P/B-ratio’s imperfections as a sole investment criterion, we still believe that the measure could do fairly well in a large sample. That is, given our investment universe of 875 stocks, a portfolio representing the stocks with the lowest P/B-ratios should outperform a portfolio with stocks trading at the highest P/B levels.

Looking at the results of equally weighted portfolios, the portfolio with the highest P/B-stocks (5th quintile) will yield the lowest returns for every year in the sample. The 1st quintile portfolio is less consistent, being the best performing portfolio in only two years out of six. That said, the 1st quintile will outperform the 5th quintile in each single year, implying that a long/short strategy would do well. Looking at the returns in 1999, a long/short strategy would take a considerable hit. As expected, this investment strategy would not be successful in this year, since many ‘growth’ stocks, with typically high P/B-ratios, hit the roof on the stock market. Perhaps this illustrates that a dynamic measure, which leans more heavily on P/B in bear-markets and less in bull markets, could do better as a stock selection criteria.

IBES Earnings Forecast (for FY1)

IBES (Institutional Brokers Estimation System) is a service that collects future earnings forecasts of major security analysts for US companies. In our model, we used 1 year EPS forecast, which is the most recent earnings forecast for the next financial year.

We used the same method to see the generating power of excess alpha.

This factor gave a consistent negative alpha in portfolio 5 (with the exception of 2003). With these results, we can assume that this factor has significant power to generate alpha when shorting low scoring stocks.

We can also see little bit weaker than portfolio 5 but still acceptable consistency in portfolio 2. Although portfolio 2 has relatively small excess return, it has positive return for 5 out of the 6 years. Also, only negative return in 2002 has relatively small differential from market return. Therefore, we believe that we can generate positive alpha by longing portfolio 2. One thing we should note is portfolio 2 has relatively small beta, 0.739. This means we have to panelize alpha from this portfolio.

From these result, we can conclude that it is highly possible to get reasonable alpha by combination of buying portfolio 2 and shorting portfolio 5.

MULTIFACTOR MODEL

Next, we developed a model where stocks are selected based on their combined performance in the earnings yield, the B/P and the IBES EPS forecast factors. This proceeds as follows:

1. Depending on how attractive a portfolio’s characteristics is, we assign it a score ranging from -5 to +5, where a negative score implies a feasible ‘short portfolio’ and a positive score is attributed a portfolio suitable for a long position.

2. For each month, every stock in a portfolio is assigned a score corresponding to the portfolio score determined in step (1). That is, if B/P (1) portfolio was given a score of +3, each of the stocks in this quintile for a given month will be attributed a score of +3 in this month.

3. All scores assigned to a given stock in a month are summed.

Based on the total scores from step (3), we build new quintiles for each month.

Hopefully, a trading strategy based on quintiles constructed using these multiple factors will yield a performance superior to a strategy based on any of the individual factors.

SCORING OF FACTORS

We assigned the P/B (1) a neutral score, i.e. zero, since it was far from consistently the best performing quintile. The P/B (5) portfolio however, yielded the lowest returns for all of the six years in the sample and was therefore attributed a -2 score.

B/P Heatmap

[pic]

Apart from year 2003, we find that the earnings-yield factor is doing rather well, and hence we assign the EarY (1) portfolio a score of +4 and the EarY (5) a score of -4.

EarY Heatmap

[pic]

With regards to the EPS factor, we do not put much value in the EPS (1) portfolio’s performance, and assign it a neutral score (0). The EPS (5) is awarded a -2 score, due to low returns in 2000 – 2002.

EPS Heatmap

[pic]

[pic]

RESULTS, MULTIFACTOR MODEL

Running our multifactor model with the above assigned scores for each of the three factors yielded the following result:

Multifactor Heatmap

[pic]

[pic]

As seen, using multiple factors leads to more consistency in returns. Over all the result is satisfactory, with the exception of 2003 where a short/long strategy would be rather undesirable. Also, note that the quintile 5 portfolio now has a negative, albeit insignificant, alpha.

DYNAMIC WEIGHT MODEL

Next, we modified the model which we made to implement dynamic weights factor. Main assumption we made is our subjective weights given to three factors should be changed along with economic perception by the market. We choose yield curve shape as indicator of economic forecast.

INDICATOR FOR DYNAMIC WEIGHTS

We categorized yield curve shape as strong positive, flat, and negative. Under the condition that there is limited risk of future inflation, we can assume a strongly positive yield curve slope represents higher expected growth, whereas a negative slope represents economic recession. Flat represents neutral stance of the market for future economic growth.

Yield curve shape is defined by difference in YTM rates between 10 year and 1 year US government bonds. We set yield curve shape as strong positive when this difference is larger than 1.5%, flat when the difference is between 0% to 1.5%, and negative when less than 0%. The distribution is as follows.

[pic][pic]

GIVING SUBJECTIVE WEIGHTS FOR EACH ECONOMIC FORECAST

First, we set subjective weight which we used in fixed weight model as reference point to figure out what effects might occur in weights when economic forecast changes. So, we set those represents for flat situation. Next we estimate what will happen in weights for three factors when the market expects strong future growth.

Strong growth

For the Forecasted EPS and Earnings Yield, we estimated that nothing will happen for buy portfolio because future expected growth will not change the market perception for buy portfolio, which already has the highest Earnings factor. But if we see sell portfolio, which has lowest earnings, the market expectation will change. In bullish market, investors have tendency to prefer value, or low EPS, stocks because these stocks have potential to improve business results. Therefore, we eliminated or reduce negative weights for these two factors. For price to book ratio, we expect same tendency because high price to book ratio implies future cash flows explains most of the stock price and vise versa.

Recession

For the Forecasted EPS and Earnings Yield, we estimated that nothing will happen for SELL portfolio because future expected recession will not change the market perception for sell portfolio, which already has the lowest earnings factor. But if we see buy portfolio, which has highest earnings, the market expectation will change. In bearish market, investors will loose confidence about future earnings. We can expect that this tendency affects more adversely to those which have highest earnings level. Therefore, we eliminated or reduce positive weights for these two factors. For price to book ratio, we expect opposite behavior and reduce negative weight for sell portfolio because low price to book ratio implies assets, less sensitive than future earnings, explains most of the stock price.

The dynamic weights are summarized in following table.

The reason we didn’t use regression nor optimizer is that, given limited length of periods and complication of which we want to implement, we expect that these method may loose explanatory power.

RESULTS

Following are our summarized results.

Dynamic Multifactor Model Heatmap

[pic]

By implementing dynamic weighting factors, we can improve performance for both buy and sell portfolio. (Please note in sell portfolio, decline in return is favorable.) Although we still have overall problem in 2003, we can observe consistent improvement for buy portfolio. Also, we see less consistent, but still favorable improvement in sell portfolio. By combining these, we can see significant improvement in cumulative return both for last two years and five years.

These results imply that our way of modification for factor weights have basically collect direction. Also, by carefully examining migration of individual stocks will provide us whether our reasoning is collect or not.

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

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

Google Online Preview   Download