Sector Model Stock Selection Service

Transcription

1 C olumbine C apital S e r v I c e s, I n c. Sector Model Stock Selection Service Annotated Presentation 2007

2 Copyright 2007 by Columbine Capital Services, Inc. All rights reserved. Columbine Capital Services, Inc. Two North Cascade Avenue Suite 450 Colorado Springs, CO Telephone: Fax: URL:

3 Notes on Model Results Columbine s models are intended to aid investment professionals familiar with industry practices and statistical tools at the CFA level in screening for stocks demonstrating historically successful measures of value and momentum. The results presented in this document are intended to illustrate the statistical characteristics of those models. These results were achieved by the retroactive application of models that were designed with the benefit of hindsight. When evaluating the performance of such models it is important to consider the following: The results reported here are hypothetical. Unless otherwise noted, they were compiled after the end of the period specified and do not represent decisions made by Columbine Capital Services during that time. As such, these results do not reflect the impact that any material market or economic factors might have had on Columbine's application of these models if they actually had been used to manage client assets during the periods presented here. These results do not represent actual trading using client assets. They should not be considered indicative of the investment skill of Columbine Capital Services, Inc. Columbine has never managed client funds according to the strategies depicted in this document, nor does it offer investment management services based on these strategies to investors. Clients for Columbine's research services actually had investment results that were materially different from the results portrayed here. The performance of past rankings does not assure the profitability or utility of future rankings. Used in isolation, these models could generate frequent relative and absolute losses; annual portfolio turnover could exceed 300%. Concentrations in the stocks of particular economic sectors or industry groups would be common. Unless otherwise noted, all returns are calculated on a time-weighted basis, using monthly valuation, based on equal-weighted deciles and universes. Returns include reinvested dividends (total return) and are presented gross of brokerage commissions, market impact, or other expenses of trading. Subscription fees for Columbine s services are not included in the performance calculations; a client s actual return would be reduced by the incorporation of those fees. The effect of fees and expenses on performance will vary with the relative size of the fee and account performance. For example, assume assets under management of $500 million, and an annual rate of return of 10.0%. The compound effect of an annual subscription fee of $25,000 per year over a ten-year period would reduce that annual rate of return by 0.34 basis points to 9.997%. We base our computations on data from commercial sources that we believe to be reliable, but we cannot guarantee the accuracy of that data. Decile-by-decile results for all Columbine models during the periods reported on here (or for any period) are available on request. A complete history of every ranking made by Columbine Capital Services, Inc. is available for inspection in our offices. Notes on Results ver.03-1.doc 02/13/03

4

5 Agenda Introducing Columbine Capital Services Who we are What we do Service overview Model characteristics Coverage & delivery Performance How we can add value in your process Applications for investment decision-support This presentation is broken up into three main sections: Introducing Columbine Who we are and what we do. Why some of the industry s top institutional investors have chosen to work with Columbine. A overview of the service How are the Columbine rankings computed, what can you expect from the stocks ranked buy and sell, and how successful have the Columbine forecasts been? How we can add value in your process Practical applications for the Columbine rankings in support of your investment decision process. Sector Model Service 1

6 Who Is Columbine? Independent Research Provider Offices in Colorado Springs, CO Founded in 1976 by John Brush Institutional clientele from US, UK, Canada Employee-owned corporation 3 principals, all 20+ years in industry Support staff of five Objective research Not broker-dealers, do not run money Columbine Capital Services, Inc. is an independent research provider with offices in Colorado Springs, Colorado. Founded in 1976 by the firm's president, John S. Brush, Ph.D., Columbine has been serving the needs of institutional investors since the early days of quantitative investing. Columbine Capital s clients include money managers, investment companies, banks and hedge funds in the United States, Canada, and the United Kingdom. The firm s three principals are: John S. Brush, PhD President Michael Anselmi, PhD Director of Research and Systems David Ament, JD Director of Marketing and Client Service Each of Columbine s principals has more than twenty years of experience in institutional quantitative research. We understand your business and we appreciate your problems. As an entirely employee-owned company we offer you objective analysis, free of extraneous influences. Columbine is not a broker-dealer and it does not engage in any investment banking activities. The firm does not manage money. Sector Model Service 2

7 Top-Ranked Research Columbine Capital s rank by Investars.com as of 1-Jan-2007 Period Long Long-Short 4 years 1 st 1 st 3 years 1 st 2 nd 2 years 1 st 2 nd 1 year 1 st 5 th Firms ranking >500 stocks; performance of all stocks covered; S&P 500 benchmark Columbine Capital Services has been recognized by investment professionals as a leader in quantitative equity research for many years. Now, with the rise of independent third-party research tracking firms like Investars, that leadership can be quantified and made concrete. Columbine s equity research truly is top ranked. Every week since January of 2003 Columbine Capital has sent Investars Buy, Sell, or Hold recommendations for all the stocks in our US research coverage universe. These recommendations are driven by our institutional Sector Model, using a simple strategy algorithm to convert our 1 to 10 rankings into specific investment recommendations. Investars time-stamps our ratings and keeps track of the results, measuring the average daily gain/loss on a hypothetical $10,000 investment. The performance of the Columbine portfolios is matched up against that of the other firms submitting their research to Investars, a group that includes both small independent research firms and the major sell-side analyst research. As of the beginning of 2007, Columbine Capital s positive ratings (Buy recommendations) ranked #1 over the prior 4, 3, 2, and 1 years. The long-short portfolio (positive + negative ratings) is #1 at 4 years and has never dropped out of the top 5. These ranks are based on comparisons of Columbine s performance to that of all other firms covering more than 500 stocks. The performance is measured for all stocks covered by each firm, with the positive returns benchmarked against the S&P 500 Index. For more details, visit Sector Model Service 3

8 What Columbine Does Search for new sources of alpha Original quantitative research Build alpha-forecasting models Unique, innovative methodology Apply models to rank stocks worldwide Weekly rankings for 28,000+ companies Deliver rankings to subscribers Internet & hard copy reports Work with clients Integrate our rankings into your process All of Columbine s services spring from our extensive R&D. We have been conducting original research into the sources of alpha (active return) for more than two decades. What we learn about alpha sources finds expression in mathematical models, each designed to forecast active return based on different return characteristics, investment styles, and country markets. Every week we gather the latest financial and market data for more than twenty thousand companies worldwide and feed that information into our models. The models output clear, unambiguous rankings (usually 1-10) that summarize each model s forecast of each company s future active return. We deliver the updated rankings from our models to our client firms in the US, the UK, and Canada. Subscribers receive updated rankings electronically and in hard copy reports. Columbine s work doesn t end with data delivery. We work closely with clients to help them integrate our rankings into their own processes. Some of the world s most successful professional money managers support their decision-making process with quantitative equity rankings from Columbine. Our clients include money managers, banks, mutual funds, insurance companies, hedge funds, and plan sponsors. Sector Model Service 4

9 Sector Model Service Sector-specific multifactor stock analysis Appropriate for a wide variety of investment products Bogey: Columbine 1500 Universe Published since 2001 The Columbine Sector Model is a Stock Selection Model designed to forecast alphas in all kinds of stocks. The model s benchmark is the Columbine 1500 Universe. The Sector Model capitalizes on the observed fact that companies in different economic sectors often have very different characteristics and need to be judged by different standards. The model accomplishes this by using sector-specific analytical frameworks to arrive at its rankings. The Sector Model is appropriate for use by portfolio managers whose results are judged against a general equity benchmark, or a particular sector peer group. We introduced the Sector Model in 2001, so much of the historical performance reported here is based on backtests in our research database, not on published rankings. We tried to simulate as closely as possible the environment that would have faced an investor at the time. For example, throughout the historical database individual stocks were assigned to the particular sector they were classified in at that time, regardless of their current designation. Sector Model Service 5

10 Sector Service Results Top decile alpha vs. Columbine 1500 through Calendar Year -1.2% 3 Years 4.3% 5 Years 2.8% 10 Years* 4.9% Since Inception 3.7% Published rankings in Columbine 1500 Universe since 12/31/00 (*includes backtest) Annualized return, monthly rebalancing, no costs Although developed in a historical database, the Sector Model has produced outof-sample results consistent with its development testing. Here are the returns generated by the Sector Model s top decile stocks (equal-weighted) in the Columbine 1500 Universe across several different time horizons. These are annualized alphas vs. the universe at each time period. We began publishing the Sector Model rankings at the beginning of 2001, so the ten-year return figure includes backtest rankings. At the beginning of every month we measure each stock s attractiveness with the Sector Model and rank-order all the issues into deciles (equal 10% groupings). Stocks in the top 10% (ranked 1) are most likely to outperform the market, and those in the bottom 10% (ranked 10) are most likely to underperform. At each month-end we compute the average return (equal-weighted) for the stocks in each decile and the return of the equal-weighted Columbine 1500 Universe to determine the alpha (excess over the benchmark return). We then link these monthly active returns for the entire period and calculate each decile s annualized rate of return. No transactions costs are included. Please see the Notes on Model Results. Sector Model Service 6

11 Model Creation Process Gradient maximization optimization Iterative, multiperiod portfolio simulations Monthly rebalancing past 10 years Stepwise changes to candidate model Utilizes costs, other constraints Significant advantages Builds in portfolio risk considerations Exploits non-linear factors Unique process yields unique models Annual re-optimization keeps model fresh At Columbine Capital, our principal tool for designing multifactor alphaforecasting models is an optimization technique from the world of operations research called gradient maximization (grad max) 1. This is a search process that iteratively performs thousands of portfolio simulations across years of market data. With each iteration we systematically vary the mix of factors and factor weightings of the candidate stock ranking models in discrete steps, searching for the optimal model for a given investment strategy. The goal (objective function) of our grad max optimization process is to maximize the IR (information ratio) of the held portfolio. Because the simulations can include the effects of transactions costs and other realworld constraints appropriate to the desired investment objective, the resulting model incorporates portfolio risk considerations. Grad max s focus on the held portfolio easily exploits any beneficial nonlinearities of the factors. The unique nature of the Columbine process leads to unique models, avoiding the me too nature of so much quant research. Every year we re-optimize the model over the previous ten years of data. 1 For more detailed information on the gradient maximization methodology, we recommend: Brush, J.S., and V.K. Schock "Gradient Maximization: An Integrated Return/Risk Portfolio Construction Procedure." Journal of Portfolio Management, vol. 21, no. 4 (Summer): (Reprints are available from Columbine) Sector Model Service 7

12 Can One Size Fit All? Sector Average Characteristics as of December 29, 2006 Characteristic P/E P/Book Val Dividend Yield Debt-to-Capital ROE Beta (vs SPX) Info Tech % 9% 18.7% 1.40 Energy % 19% 28.1% 0.83 Source: Thomson Baseline The table above sets out the average characteristics for the stocks of two different economic sectors: Information Technology and Energy. (These figures are from the Thomson Baseline system as of the December 29, 2006 close.) Not surprisingly, the two sectors are very far apart on most metrics. Some of this difference is accounted for by investors contemporary view of these sectors, but a lot of the variation is systemic IT and Energy simply are very different kinds of companies. Given the disparity between the characteristics of these two sectors it s difficult to imagine how a single ranking model could fit them both. The truth is, it can t. General purpose quantitative models are, of necessity, compromises they attempt to forecast alphas for all stocks based on a single combination of factors that isn t optimal for any particular subset. Such a one size fits all approach may succeed most of the time, but when a particular sector dominates the market (e.g., the tech stocks in the 1990s) the compromise approach isn t able to cope. Sector Model Service 8

13 Tailor-made Models Momentum Factors Price Momentum Estimate Revision Earnings Change Earnings Growth Earnings Surprise Sector- Specific Models (optimal weighting of input factor rankings) Valuation Factors Book Value/P Trailing EPS/P Dividend Yield Estimated EPS/P Cash Flow/P Stock s Sector Model Score Sort entire universe and divide into deciles Stock rankings from the Sector Model are a synthesis of the inputs from the ten individual factors listed above. The Sector Model for each economic sector combines each stock s rankings from these factors in a linear equation of ten terms: Sector raw score = B 1 f 1 + B 2 f B 10 f 10 Where B is the sector-specific sensitivity of a stock s future active return to the value of its corresponding factor f. We then sort all the stocks on their Sector Model scores and divide them into deciles. The most attractive 10% are ranked 1 and the worst are ranked 10. Since the momentum and valuation factor rankings that go into the Sector Model are relative to all stocks, the resulting Sector Model rankings are predictive of active return relative to all stocks, not just relative to that particular sector. The rankings can be used successfully to create unconstrained portfolios (no control of sector weightings) or to generate a sector-neutral portfolio (sector weightings match the universe). Sector Model Service 9

14 Some Are Very Stable Relative factor weightings: Information Technology Sector Valuation Momentum '90 '91 '92 '93 '94 '95 '96 '97 '98 '99 '00 '01 '02 '03 '04 '05 '06 Optimization year The factor recipe for the Sector Model s component models are not static. We re-optimize each sector s factor weighting structure annually using our Gradient Maximization technology. The sample period for each year s re-optimization is the most recent ten years of historical data. The optimal blend of factors for some sectors has been quite stable over time. For example, the optimal blends for the Information Technology stocks (displayed above) have heavily favored momentum (green) over valuation (blue) since Sector Model Service 10

15 Others Change Relative factor weightings: Utilities Sector Valuation Momentum '90 '91 '92 '93 '94 '95 '96 '97 '98 '99 '00 '01 '02 '03 '04 '05 '06 Optimization year Other sectors have proven more variable. In the Utility stocks the optimal blend has changed dramatically over the past years. Part of this variation is no doubt driven by the fact that many formerly regulated utilities morphed into technology companies during these years. Since our historical database carries companies in the sector they belonged to at the time, this evolution will be seen in the data. Sector Model Service 11

16 Updates & Delivery Model rankings updated weekly Data as of Friday close Updated rankings available Saturday AM Delivery options Electronic data files Download from Columbine Web Site Automatic delivery via or FTP FactSet system Hard copy Columbine Research Book We compute the Sector Model rankings every weekend based on Friday closing prices and fundamentals. Updated ranking data typically are available on Saturday morning. (If you re ever in doubt about the currency of our information, check the Update Status display on the Columbine web site which reports the date of the last ranking update.) Sector Model rankings are available electronically through the Internet as data files that can be loaded into Excel, Access, or other applications on your system. We can automatically deliver your data files each weekend via or FTP, or you can download files yourself from a password-protected client page on Columbine's web site. The web site delivery option is particularly useful for clients who often work from home or while traveling. Columbine subscribers who are FactSet users can access Sector Model rankings through the Universal Screening and Data Downloading menus. If you prefer paper to electronic information we can send you hard copy Research Book reports of the Sector Model rankings via overnight express for delivery on Tuesday. These Research Book reports also are available as PDF files for electronic delivery over the weekend. Sector Model Service 12

17 Stable Rankings Average change in Sector Model ranking (deciles) Months after initial ranking Test universe: Columbine 1500 Universe. Test period: Model version: How stable are the Sector Model s rankings? Do you run the risk of buying a stock ranked 1 this week only to find that its ranking has changed to a sell the next? The Sector Model s rankings do change, of course, but the changes usually are gradual and progressive. The graph above sets out the Sector Model s typical ranking change over four different time periods: 1, 3, 6, and 12 months. In one month the average change is just one decile rank; even over the course of a year the typical stock will change its ranking by only two or three decile ranks. Sector Model Service 13

18 The Coverage You Need Columbine Total US Universe: ±6,000 issues All NYSE- and AMEX-listed common stocks OTC issues priced $1.00 or higher Benchmark: Columbine 1500 Universe Institutional Household Names list S&P 500, plus 1,000 stocks selected on: Trading volume ($) over past year Institutional ownership Analyst coverage We apply the Sector Model to every stock in our domestic Columbine Total US Universe; this includes all NYSE- and AMEX-listed common stocks, along with OTC issues priced at $1 or more. The Sector Model s rankings are benchmarked to the Columbine 1500 Universe, our own institutional household names list of companies. Our Columbine 1500 Universe consists of all the S & P 500 companies, plus another 1000 names selected from the top 2000 companies by market cap. We reconstruct the Columbine 1500 Universe annually, based on a three-factor screen: 1) institutional ownership; 2) dollar trading volume over the past year; and 3) analyst coverage. Sector Model rankings are available for several standard report universes: our own Columbine Total US, or Columbine 1500 Universes; the S&P/BARRA 400, 500 or 600 stocks; or the Russell 1000, 2000 or 3000 stocks. We can create custom reports for specialized stock universes based on characteristics such as market cap. Sector Model Service 14

19 Rank: Easy to Interpret Average alpha (annualized) by Sector Model rank 11.1% 7.3% 4.2% 2.0% 1.2% -0.3% -1.7% -5.3% -5.8% -13.1% Test universe: Columbine 1500 Universe, test period: Model version: month holding, annualized, no costs It s easy to interpret and use the Sector Model rankings: the lower the ranking number the greater a stock s expected alpha. The cross-over point between positive and negative expected alpha falls in the middle ranks, and the higher ranks produce the largest negative alphas. The chart above sets out the average alpha versus the universe (annualized) associated with each rank (equal-weighted decile) of the Sector Model in the Columbine 1500 Universe stocks over a thirty-five-year test period. The chart s results are based on one-month holding periods and are reported gross of transactions costs. Generally, we consider stocks in the model s top 10% (ranked 1) to be buys, and those in the bottom 10% (ranked 10) to be sells. This does not, however, imply that you should sell a stock just because its ranking slipped below 1 doing so would lead to unacceptably high levels of turnover. Neither should you wait until the ranking dropped all the way to 10 before selling that would sacrifice too much return. In our section on model applications we ll show you how to choose buy/hold/sell strategies to fit your fund s specific requirements. Sector Model Service 15

20 12% 10% 8% 6% 4% 2% 0% -2% -4% -6% -8% Long-term Utility Average alpha (un-annualized) by Sector Model rank Holding period (months) Rank: 1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th Test universe: Columbine 1500 Universe; test period: Model version: Un-annualized, no costs Long-term predictive ability is vital if model rankings are to be useful in institutional portfolio strategies. The graph above illustrates the average alpha versus the universe associated with the Sector Model s ten ranks at holding periods of 1, 3, 6, 12, 24, and 36 months (un-annualized). These are un-rebalanced, buyand-hold results, gross of transactions costs. Stocks identified as buys (ranked 1) by the Sector Model tend to generate significant positive alpha for as long as three years. It is this ability to identify stocks that keep generating alpha month-after-month, year-after-year, that gives the model its utility in institutional portfolio management. The sells (ranked 10) look like the negative alpha analogue of the 1st rank returns. The results for the interior ranks (2-9) pretty much line up in a nice monotonic sequence. Sector Model Service 16

21 Works in all Sectors Average Sector Model top- minus bottom-decile spread by sector Cons Staples 11.8% Cons Discretionary 19.4% Health Care Materials Industrials Info Technology 12.8% 10.5% 14.6% 13.9% Energy 18.1% Telecom Svcs Utilities 11.5% 13.5% Financials 18.4% Test universe: Columbine 1500 Universe, test period: Model version: month holding period composite, annualized, no costs The Sector Model is predictive in all economic segments. Although there is variation from sector to sector, the Sector Model does an excellent job of discriminating between buys and sells in every economic sector. The chart above sets out the Sector Model s average top- minus bottom-decile spread in each of the ten GICS economic sectors within the Columbine 1500 Universe over a thirty-five year test period. This study treated each sector as a separate universe of stocks and used the Sector Model to rank those stocks relative to the sector. We report a composite of the average spread results at holding periods of 1, 3, 6, and 12 months. The graph bars represent the average spread return in each sector between the issues ranked 1, and the issues ranked 10. These are annualized, equal-weighted results gross of transactions costs. Sector Model Service 17

22 Annual Active Return Sector Model Service Annual top decile alpha 40% 35% 30% 25% 20% 15% 10% 5% 0% -5% Backtest Published '86'87'88'89'90'91'92'93'94'95'96'97'98'99'00'01'02'03'04'05'06 Results in the Columbine 1500 Universe, monthly rebalancing, no costs Here is the history of the Sector Model s top decile year-by-year results. The bars represent the annual alphas of the stocks the model ranked 1 ( buys ). These results are excesses of the equal-weighted decile return over the equal-weighted Columbine 1500 Universe return, based on monthly rebalancing, and are gross of transactions costs. We introduced the Columbine Sector Model at the end of 2000, so the results shown prior to 2001 are based on backtests of the model in our research universe, not published rankings. The Sector Model varies in its effectiveness from year-to-year, just as any other quantitative tool. The published rankings (2001-to date) have been lower than the backtest results, largely because of the dominance of value-oriented factors during much of this period. In such markets the Sector Model tends to look weak at the one-month holding periods measured here, though the stocks it selects still may do well at the longer time horizons the model was designed for. Please see the Notes on Model Results. Sector Model Service 18

23 Applications Generate buy ideas Stocks ranked 1 (top 10%) are buys Analyze your own buy candidates Sell discipline Stocks ranked 10 are absolute sells Create your own optimal sell strategy Alpha estimate for optimizer Foundation of stock selection system Here are some of the ways our clients apply the rankings from Columbine s Sector Model: Many managers use the rankings simply as a source of new buy ideas. Issues ranked 1 are current buys. Starting your own analysis with a short list of highpotential issues saves time and can improve your portfolio s return. Let the model s rankings act as an objective, unambiguous sounding board for analyzing buy candidates suggested by other research. If a candidate is poorly ranked by our model, you may want to re-examine the original rationale behind its purchase. Setting up an effective sell discipline can be just as valuable as finding strong stocks to buy. Always sell stocks ranked 10 (10th decile); the future prospects for these issues are uniformly bad. We can help you determine optimal sell trigger points for a particular client or product that maximize return without exceeding the appropriate turnover level. Firms taking a more quantitative approach often use Columbine model rankings as a source of alpha forecasts for input into a portfolio optimizer. Rankings from one or more of our models can provide the foundation for the entire stock selection function in your equity management process. Sector Model Service 19

24 Annual alpha (net) 8% 6% 4% 2% Sector Strategy Map 1 Buy stocks ranked 1, hold through: Unconstrained Sector-neutral 9 9 0% Average holding period (months) Test universe: Columbine 1500 Universe, test period: Model Version: Monthly rebalancing, net of transactions costs The graph above is a Strategy Map for the Sector Model. With it you can determine the particular portfolio strategy that is most appropriate for your own funds or investment clients. The two numbered series illustrate the alphas and holding period characteristics generated by portfolios run under nine different buy/hold/sell strategies across a thirty-five year test period. Each of the strategies assumed a 150-stock portfolio, equally weighted, rebalanced monthly. The green line sets out the results for unconstrained portfolios run without regard to sector balance. The blue line illustrates results for sector-neutral portfolios where the sector distribution always matched that of the universe. All returns are net of transactions costs based on a variable costs algorithm on average about 1% round-trip. The starting portfolio for all the strategies consisted of those issues ranked 1 (1st decile) by the Sector Model at the beginning of the simulation. What varied from strategy to strategy was the model ranking level that would change a hold into a sell. Stocks that dropped below their strategy s hold level were sold and replaced with current 1-ranked issues. The strategy numbers on the chart correspond to the lowest-ranked holding tolerated in each strategy. For example, strategy 1 held only stocks ranked 1. Strategy 2 held stocks ranked 1 or 2; Strategy 3 held issues ranked 1, 2, or 3, and so on. Sector Model Service 20

25 Typical Portfolio Strategy Strategy 3 Unconstrained Buy stocks ranked 1 Hold as long as ranked 1, 2, or 3 Sell when rank drops to 4 or lower Portfolio Characteristics: Alpha: 6.7% (annualized) Beta: 0.95 Information ratio: 0.93 Average annual turnover: 90.9% Hit rate: 89% years; 70% quarters Test universe: Columbine 1500 Universe, test period: Model Version: Monthly rebalancing, net of transactions costs Strategy 3 represents an application of the Sector Model that would probably mesh well with the approach followed by many institutional portfolio managers. Under this strategy the candidate buy list is made up of the stocks currently ranked 1 by the Sector Model. Portfolio components are held as long as they ranked 3 or better. When any holding drops to a Sector Model rank of 4 or lower it is sold and replaced with a 1-ranked company. We simulated this strategy in the stocks of our Columbine 1500 Universe over the years We rebalanced the portfolios monthly and we accounted for transactions costs by using a variable costs algorithm that recognizes the varying levels of liquidity presented by different types of stocks and markets. The strategy was not constrained for sector balance we let the model make sector bets in the portfolios. The result was a monthly alpha of 6.7% (annualized) from portfolios that were less risky than the market (beta of 0.93). The strategy s information ratio (mean monthly active return divided by the standard deviation of the active returns) was just over 0.9. Annual portfolio turnover was less than 100%, and the strategy beat the universe in 89% of the years and 70% of the quarters. Sector Model Service 21

26 Strategy 3 Results Cumulative monthly return net of transactions costs 7 6 Strategy: 20.0% Universe: 13.2% 5 Return (ln) '71 '76 '81 '86 '91 '96 '01 '06 Test universe: Columbine 1500 Universe, test period: Model Version: Monthly rebalancing, net of transactions costs Here is the cumulative monthly return series generated by the Sector Model s strategy 3, along with the returns for the universe from which the portfolios were selected (Columbine 1500 Universe). These are absolute total returns (dividends re-invested) based on equal-weighted portfolios of 150 stocks and an equalweighted universe. The annualized rates of return for both series, based on their terminal values, are reported at the top of the chart. It is easy to see that the Sector strategy systematically outperformed the universe over the course of the study. There are ups and downs, but clearly the Sector strategy consistently added value. While this is a backtest, it is important to remember that the 2006 version of the Sector Model used for this simulation was fit in just the ten years of data for 1996 through In effect, we had a twenty-five year hold-out sample. Sector Model Service 22

27 Sector Model Summary Multifactor alpha forecasting model Sector-specific structure Rankings updated weekly Variety of delivery options Unique, innovative modeling methodology Incorporates real world transaction costs Factor weightings re-optimized annually Columbine s most advanced model Wide variety of application strategies In summary, Columbine Capital s Sector Model Service is based on a multifactor stock selection model that is designed to forecast individual stock alphas as far out as three years in the future. The model applies different optimal factor weighting structures to the stocks of different economic sectors. Clients receive weekly stock rankings from the Sector Model, delivered electronically and in hard copy. The innovative gradient maximization process that we use to create the Sector Model is unique in its focus on the held portfolio and in its ability to incorporate transactions costs and other real world constraints directly into the model creation process. We re-optimize the Sector Model weighting structure annually to keep the model current. The Sector Model Service is Columbine s latest and most advanced stock selection model. Its sector-specific approach is the culmination of everything we have learned in three decades of quantitative research. Clients make use of the Sector Model rankings in a wide variety of ways, ranging from simply using it as a source of buy ideas to making it the foundation of their entire stock selection system. Sector Model Service 23

28 A Manager s s Resource Save time Spend your own analysis more profitably Save resources Creating your own systems = $$$$ Improve return Proven forecasting ability Columbine s Sector Model can help you: Save time Use the model to screen a large universe of stocks down to a smaller list of strong buy candidates. This lets you spend your time and analytical talents on choosing among those issues that have already been determined to be attractive. Save resources Many of our clients could create a multifactor alpha forecasting model of their own; some already had one when they hired us. But building a model that can approach the long-term predictive power and stability of the Sector Model requires an expenditure of resources greatly in excess of the model s annual subscription cost. The make vs. buy decision is clear. Improve return In the highly competitive business of investment management small differences in return can have big payoffs. With the Sector Model you have a state-of-the art, active return forecasting tool that is continuously updated and improved to keep it adding maximum value in your portfolios. Sector Model Service 24

29

30 Columbine Factors Columbine utilizes thirteen individual return and risk factors in creating the Columbine Component Models and our Stock Selection Models. Each multifactor model has a different recipe of these thirteen factors, and not every factor is used in every model. Columbine s criteria for selecting these particular factors are two-fold: First, each factor has to make economic sense; simply exhibiting a correlation with future return is not sufficient without a sensible rationale. Second, each factor has to demonstrate significant predictive power on its own. In the case of the three risk factors that predictive power is more related to the volatility of future return. Ranking convention: Deciles (1-10). For return factors a ranking of 1 (top 10%) indicates most likely to outperform and 10 (bottom 10%) indicates most likely to underperform. For risk factors a 1 ranking indicates the lowest risk, while 10 is the highest risk. Return Factors BOOK VALUE The book value-to-price ratio. Defined as: current book value divided by current stock price. CASH FLOW The cash flow-to-price ratio. Defined as: trailing four-quarter EPS, minus trailing twelve-month dividends, plus depreciation, divided by the current stock price. COLUMBINE ALPHA Columbine's objective measure of each stock's price momentum. Based on a unique, volatility- and risk-adjusted model, the Columbine Alpha Factor forecasts each issue's probable performance over the next six to twelve months. Defined as: the alpha estimate resulting from a generalized least squares regression of weekly (Fridayto-Friday) stock price changes for the past twelve months against weekly changes in the S & P 500 Composite Index for the same period. We apply a proprietary weighting structure to the stock price change and market change data before the regression is run. DIVIDEND YIELD The ratio of a stock's dividends paid during the past year to its current price. Defined as: trailing twelve-month dividends, divided by the current stock price. EARNINGS CHANGE Short-term change in EPS. Defined as: current trailing twelve-month EPS, minus trailing twelve-month EPS one quarter ago, divided by current stock price. EARNINGS GROWTH Long-term change in EPS. Defined as: current trailing twelve-month EPS, minus trailing twelve-month EPS two years ago, divided by current stock price. Columbine Factors.doc 2-Feb-07

31 EARNINGS SURPRISE A comparison of a company s reported earnings to the latest consensus estimate. Defined as: actual reported EPS for the fiscal quarter just completed, minus the latest median EPS estimate for the same fiscal quarter, divided by one plus the latest median EPS estimate for the quarter. Stocks ranked 1 have had the largest positive earnings surprises; those ranked 10 have had the largest negative surprises (disappointments). If the reported and estimate figures are the same, the stock is ranked a neutral 5. If the company s last earnings announcement is more than 30 days old the stock is ranked a neutral 6. ESTIMATED EARNINGS YIELD An earnings yield based on the consensus EPS estimate. Defined as: median estimated EPS for the next four quarters, divided by the current stock price. ESTIMATE REVISION A multifaceted measure of revisions during the past 60 days to sell-side analysts estimates of each company s FY1 earnings. Double weight is given to changes in the last 30 days. If a company is more than six months into its fiscal year a blend of FY1 and FY2 data is used. The factor ranking is a synthesis of three individual components: Analyst Agreement Are the estimates changing in concert? Defined as: number of positive revisions minus the number of negative revisions, divided by the total number of estimates. Revision Confidence Are the extreme estimates changing? Defined as: change in the highest estimate plus change in the lowest estimate, divided by the current stock price. Revision Magnitude How big is the typical revision? Defined as: change in the median estimate divided by the current mean estimate. REPORTED EARNINGS YIELD The classic "value" measure based on reported EPS figures. Defined as: current trailing four-quarters EPS divided by the current stock price. BETA The standard measure of a stock's market risk. Risk Factors Defined as: the beta estimate from regressing five years of weekly stock price changes against the weekly changes in the S & P 500 Composite Index for the same five years. MARKET LIQUIDITY A measure of how difficult it may be to trade a given stock. Defined as: current stock price times the average daily trading volume over the previous full month. EARNINGS VOLATILITY A measure of how variable a company's quarterly earnings are. Defined as: the standard deviation of the reported EPS over the past twelve quarters. For More Information Please contact David Ament at , or via d.ament@columbinecap.com Columbine Factors.doc 2-Feb-07

32 Sector Model Service Results through: December, 2006 Columbine Capital Services, Inc. Columbine 1500 Universe Model description: General-purpose stock selection model that applies a different optimal multifactor model to evaluate the stocks in each of the ten S&P GICS Economic Sectors Total return (dividends re-invested), monthly rebalanced, gross of transactions costs Absolute returns Annualized historical statistical characteristics: 1971-date Month of December, 2006 Decile alpha beta sigma Sharpe R 2 Track Error Top decile 1.7% Top Bottom decile 0.1% Bottom Columbine % Cumulative Monthly Active Return vs. Columbine 1500 Universe Top decile: +4.88% annually Bottom decile: -7.24% annually Return (ln) '97 '98 '99 '00 '01 '02 '03 '04 '05 '06 Annual Active Return vs Columbine 1500 Universe (eql-wtd) Decile '97 '98 '99 '00 '01 '02 '03 '04 '05 '06 Top 13.1% 4.3% -15.0% 29.1% 8.5% 0.6% 0.6% 9.2% 5.0% -1.2% Bottom -15.1% -8.1% 6.8% -34.4% 3.5% -20.3% 25.3% -4.0% -12.3% -0.7% Annual Top-Bottom Decile Spread Return (eql-wtd) Spread 32.4% 12.7% -20.9% 88.9% -12.4% -5.5% -20.7% 13.2% 18.9% -0.9% N.B., Spread figures are computed by linking monthly top-bottom spread returns Compounded Decile Active Return vs Columbine 1500 Universe through December, 2006 Decile Month 3-mon YTD 1 yr 3 yrs* 5 yrs* 10 yrs* Incep* Top 1.4% 1.4% -1.2% -1.2% 4.3% 2.8% 4.9% 3.7% Bottom -0.2% 3.6% -0.7% -0.7% -5.8% -3.6% -7.2% -2.4% Compounded Top-Bottom Decile Spread Return through December, 2006 Spread 1.6% -2.2% -0.9% -0.9% 10.1% 0.0% 6.9% -2.2% * Annualized RoR; Incep date: 12/31/00 The results reported here are hypothetical. These results do not represent actual trading using client assets. The performance of past rankings does not assure the profitability of future rankings. Copyright, 2007 Columbine Capital Services, Inc. See the Notes on Model Results. Not for general distribution. Model Analysis 6 05-Jan-2007

33 Sector Model Service Columbine 1500 Universe Results through: December, 2006 Top- and Bottom-Decile Active Return and Spread Return Results in the Columbine 1500 universe, monthly rebalanced, gross of transactions costs Shading: 95th to 5th percentile range; Error bars = 1 standard deviation Current Mean Latest Month Year-to-Date Trailing-12 Months 10% 8% 6% 4% 2% 0% -2% -4% -6% Top Bottom Spread 60% 50% 40% 30% 20% 10% 0% -10% -20% -30% -40% Top Bottom Spread 70% 60% 50% 40% 30% 20% 10% 0% -10% -20% -30% -40% Top Bottom Spread Top Bottom Spread Top Bottom Spread Top Bottom Spread Current 1.40% -0.22% 1.61% Current -1.21% -0.71% -0.88% Current -1.21% -0.71% -0.88% Pctile Pctile Pctile Sigmas Sigmas Sigmas Current: Current period return value Pctile: Percentile rank of current value in historical series (100 = best) Sigmas: Difference of current value from historical mean in standard deviations Historical Return Series: January, 1971 through December, 2006 Monthly series Calendar year series Trailing-12 month series Top Bottom Spread Top Bottom Spread Top Bottom Spread Best 8.5% -14.3% 18.8% Best 29.1% -34.4% 88.9% Best 42.5% -39.2% 97.1% 95th 3.4% -5.2% 7.8% 95th 22.9% -27.2% 55.1% 95th 22.2% -27.3% 58.4% Mean 0.8% -1.0% 1.8% Mean 10.6% -11.2% 24.3% Mean 10.8% -11.5% 24.9% 5th -1.9% 2.4% -3.0% 5th 0.1% 4.3% -14.5% 5th -4.5% 3.5% -11.0% Worst -8.5% 32.2% -40.7% Worst -15.0% 25.3% -20.9% Worst -17.6% 62.1% -55.0% Std Dev 1.7% 3.5% 4.7% Std Dev 8.2% 10.5% 22.2% Std Dev 8.1% 10.7% 20.6% Hit Rate 73.1% 71.8% 76.2% Hit Rate 94.4% 91.7% 86.1% Hit Rate 89.8% 93.6% 90.7% t -stat t -stat t -stat Best: Best value observed in historical series 95th: 95th percentile value in historical series Mean: Mean value of historical series 5th: 5th percentile value in historical series Worst: Worst value observed in historical series Std Dev: Standard deviation of historical series Hit Rate: Percentage of periods with correct results t-stat: t-statistic of historical mean Copyright, 2007 Columbine Capital Services, Inc. See the Notes on Model Results. Not for general distribution. Model Analysis 7 05-Jan-2007

34 Sector Model Service Results Multi-Month Periods Ending December 29, 2006 The Sector Model Service capitalizes on the observed fact that companies in different economic sectors often have very different characteristics and need to be judged by different standards. The Sector Model accomplishes this by identifying each sector's distinctive response to momentum and valuation forces in the markets, and uses that information to forecast each individual stock's probable active return (alpha) for the next 1-3 years. Absolute total return in % points, based on monthly rebalancing, gross of transaction costs Equal-weighted results unless otherwise noted 4th Annualized compound return Since Oct Nov Dec Quarter year 3 years* 5 years* 10 years* 12/31/00 S & P 500 Stocks Top decile Universe Bottom decile Columbine 1500 Universe Stocks Top decile Universe Bottom decile Columbine Total US Universe Top decile Universe Bottom decile NA NA 8.59 S & P 500 Index (cap-wtd) * Includes backtest results ( ). We began publishing the Sector Model in January of The results reported here are hypothetical. These results do not represent actual trading using client assets. The performance of past rankings does not assure the profitablility of future rankings. Sector Model Service Results Update Friday, January 05, 2007

35

36 Two North Cascade Avenue Suite 450 Colorado Springs, CO Phone: Fax: Web: