Equity Research Methodology

Transcription

1 Equity Research Methodology Morningstar s Buy and Sell Rating Decision Point Methodology By Philip Guziec Morningstar Derivatives Strategist August 18, 2011

2 The financial research community understands that any forecast or prediction about a company s future harbors a degree of uncertainty. Many analysts casually compare the relative uncertainty of two companies. In fact, the very process of making an investment decision tacitly answers the question, How uncertain is it? We think the estimate of a company s intrinsic value isn t necessarily a discrete point estimate. Rather, we believe a company s fair value estimate should be perceived as a range of possible outcomes, with the decision to invest based on a sufficient margin of safety relative to the fair value estimate. We believe this approach gives investors a framework for overcoming the uncertainty inherent in the investment process. Morningstar has formalized the process of making investment decisions using uncertainty and margin of safety. Summary of the Methodology The origin of Morningstar s buy and sell rating methodology is based on a simple investment philosophy: Estimate a company s intrinsic or fair value by projecting its future cash flows, compare a security s fair value estimate with its current market price, then buy the security if it is cheap relative to its fair value estimate and sell it if it is expensive. Incorporating a sufficient margin of safety ensures that the security is indeed cheap or expensive relative to its intrinsic value. Through experience, we have found that the appropriate margin of safety varies with the company s characteristics. The more uncertain the estimate of a company s value is, the greater the margin of safety that investors should demand before purchasing its shares. Our methodology for assigning Consider Buying and Consider Selling ratings based on valuation and uncertainty quantifies and formalizes this process. The end result of our investment process is shown in the following diagram, which depicts the price/fair value ratios we require at each rating level. New Morningstar Margin of Safety and Star Rating Bands as of August 18, 2011 Price/Fair Value Star % 95% 105% 80% 135% 90% 110% 70% 155% 85% 175% 115% 80% 60% 2 Star 125% 3 Star 4 Star 50% Star Low Medium High Uncertainty Rating Very High* * Occasionally a stock s uncertainty will be too high for us to estimate, in which case we label it Extreme. 2

3 As you can see from the diagram, we assign companies to one of four uncertainty buckets. As a company s prospects become more uncertain, a greater discount to its estimated value is required before we have enough confidence to assert that the company is undervalued. Similarly, because the recommendation to sell a stock is an active decision to underweight the stock relative to an index, as uncertainty about the valuation of the stock increases, we require a greater premium to our estimate of the company s value before the stock hits our Consider Selling threshold. Development of the Framework Securities analysts are in the business of predicting a company s future, then quantifying that prediction. Quantifying uncertainty about the future is more difficult than quantifying a single fair value estimate for a company; not only is there an additional variable to estimate, but that variable requires the measurement of probabilities, a task at which human beings are notoriously unskilled. Despite these challenges, we ve structured our investment decision-making framework in a way that allows us not only to say something useful about the uncertainty of a company, but also to generate a cardinal measure of that uncertainty. If you were to ask a good analyst to place a value on a company, he or she will say something like, Its shares are worth between $45 and $55 each, or Somewhere between $25 and $75. By assigning a range to the valuation, the analyst is sharing information about the uncertainty of his or her valuation estimate and the assumptions behind that estimate. Our uncertainty rating simply formalizes this process and quantifies it within a statistical framework that fits equity markets and is also the basis of the Black-Scholes option pricing model. How Analysts Assign Uncertainty To quantify the uncertainty of their fair value estimates for the companies they cover, Morningstar analysts evaluate a range of scenarios for the companies possible outcomes. They then use these scenarios to assign a company an appropriate uncertainty rating. Our analysts assign an uncertainty rating by examining a number of different operating scenarios and the fair value estimate corresponding to each scenario. The scenario analysis process can consider many factors in the derivation of the fair value estimate, but is structured around a range of possible revenue targets, the degree of operating leverage inherent in the company s business, the degree of financial leverage it employs, and any exogenous events such as lawsuits that may contribute to the uncertainty of the value of the company s equity. After developing these scenarios, the analyst estimates a 50% prediction interval for the fair value estimate of the company and then assigns the company an uncertainty rating with Consider Buying and Consider Selling prices that best fit his or her predicted confidence interval. For example, if an analyst covering company XYZ determines that the lower-bounded scenario of the 50% prediction 3

4 interval for the fair value estimate corresponds to a price/fair value of 0.62 and the upper-bounded scenario corresponds to a price/fair value of 1.70, it would most closely align with the high uncertainty band and would be assigned a rating of high uncertainty. For companies where our analyst can t assign a 50% probability of falling within 0.50 and 1.75 times the fair value estimate, we assign an uncertainty rating of extreme, meaning that we can t reasonably handicap the value of the company. A 50% prediction interval means that half of the time, the company s true fair value will fall within the predicted range. However, it also means that 25% of the time, the company s true fair value will fall below the predicted range, which means below the 5-star or Consider Buying price, and 25% of the time, the fair value estimate will fall above the predicted range, or the 1-star or Consider Selling price. Therefore, if our fair value estimates and uncertainty ratings corresponded to true ranges of intrinsic values for every stock, both our buy and sell recommendations would outperform or underperform the market (their respective cost of capital) 75% of the time. Put another way, if our analysts do their job perfectly, both our buy and sell investment recommendations will have a 75% batting average. We selected a 50% prediction interval pragmatically because it is intuitive and easy to remember. When combined with our observed ranges of outcomes for different uncertainty ratings, a 50% interval generates Consider Buying and Consider Selling prices that make intuitive sense, based on more than a decade of experience with our rating system. How We Construct the Uncertainty Bands The Distribution of Stock Returns In determining our prediction interval for the true fair value of the company, we start with the assumption that the true fair values are log-normally distributed; however, we modify this assumption based on our empirical data set, as shown in Appendix 1. We justify this assumption with the following arguments: 3 In the long run, security prices tend to be roughly log-normally distributed. These log-normal, longrun prices are consistent with normally distributed continuously compounded returns. 3 The option market prices the uncertainty around a company s value just as our uncertainty rating does. Option prices are reasonably well described by the Black-Scholes model, which assumes normally distributed returns and a log-normal future price distribution. However, just as the option market s implied volatility skew varies implied volatility across strikes, we adjust implied volatility for the upper tail of the probability distribution to match our empirical data set. 4

5 3 We assume that, on average, prices converge to our fair value estimates in the long run. Therefore, if stock prices in aggregate follow our empirical data set, then our fair value estimates have to follow this distribution as well, unless they systematically differ from the market. 3 Empirically, for Morningstar s coverage list, the distribution of prices (period returns) over a threeyear horizon is a reasonable approximation of the midpoint between a normal Gaussian distribution and a log-normal distribution for stocks grouped by uncertainty rating. (See Appendix 1.) Log-normal probability distributions generate prediction intervals that are log-symmetric, which means that a stock has the same chance of falling by half as it does of doubling. Similarly, a variable would have the same chance of falling to one fourth of its original value as it would of rising by 4 times. Perhaps a more intuitive way for stock investors to think about log-normally distributed prices is that it takes the same return to go from the Consider Buying price to the fair value estimate as it does to go from the fair value estimate to the Consider Selling price over a long horizon. Given our empirical data, we formally define the probability distribution for our fair value estimates as midway between the normal distribution, where the chance of falling by half or rising by 50% is the same, and the log-normal distribution, where the changes of falling in half or doubling are the same. We use a midpoint definition rather than assigning a distribution based on a particular sample data set because we believe any single data set would require periodic adjustment as the sample set changed over time, and because we believe the outcomes of the particular data set would imply false precision regarding our estimate of the probability distribution. Grouping by Uncertainty The probability distribution of a company s true fair value is arbitrarily shaped, based on the company s characteristics and the fundamentally derived distribution of the potential outcomes. By forcing all companies into a standard framework, we thereby lose some information. However, the complexity of estimating and tracking a different distribution shape of probable outcomes for every company we cover would be overwhelming to manage and would probably increase the perceived precision more than the accuracy. We use similar logic for grouping our companies into buckets that we have labeled low, medium, high, and very high uncertainty, because a separate value of uncertainty for each company would add complexity and imply a greater level of precision than we think is possible. In the aggregate, placing all companies into a standard framework is an approximate but effective compromise for grouping companies into buckets based on uncertainty. Time Horizon To estimate a prediction interval for the fair value of a stock, we need to assume a time horizon for that interval. Morningstar assumes a stock will converge to its fair value estimate within three years, a period that we think best represents the time constant for most fundamental factors that affect the earning power of companies. Recessions (other than the Great Depression) are over 5

6 within three years. Merger integrations are typically complete within three years. Operational changes in a company should be complete and most new products can be developed within three years. Anecdotally, economic reality usually catches up with accounting games in three years. Empirically, stock prices generally converge to our fair value estimate within three years and typically sooner, unless the fair value estimate is in error. If we assume that stock prices converge to their fair value within three years, we can infer that, on average, adjusting for expected return, a stock price three years in the future is an accurate estimate of the true fair value today, or that the stock price today is an accurate estimate, on average, of the true fair value from three years ago. Fair Value Estimate of Uncertainty The uncertainty rating can be interpreted as a fair value estimate of the implied volatility of threeyear options on the underlying company. Morningstar s uncertainty rating describes a standard price distribution at a known time horizon in the future. Therefore, each uncertainty rating represents a fundamentally derived estimate of the implied volatility skew, which is the standard deviation of returns underlying the log-normal volatility framework over that horizon, adjusted for the relationship of the ending stock price to the beginning stock price. The fair value estimate of the stock price represents the median value of the distribution, and the 25th and 75th percentiles represent our 5-star and 1-star prices. In addition, the 4-star and 2-star ratings represent the 42nd and 58th percentile of the probability distribution for the fair value estimate three years in the future, values that were chosen by optimizing the historical returns from a buy/sell trading strategy based on these ratings. The relationship between these breakpoints is shown graphically below. Fair Value Estimate of Uncertainty Probability 7 Uncertainty is the confidence interval around equity value 5 Fair Value Estimate 25th Percentile 75th Percentile Company Value 6

7 As the uncertainty (implied volatility) increases, the width of the distribution of potential outcomes increases, increasing the spread between the prices at which we would grant a buy or sell rating. This increase in the width of the distribution is shown graphically, as follows. Fair Value Estimate of Uncertainty 1 Probability 1 Price ($) Determination of the Bands and Breakpoints We determined that four bands provided sufficient resolution to differentiate among the different levels of uncertainty in our coverage universe, but that any further differentiation would amount to false precision. We calibrated the price/fair value ratios that determine the breakpoints within the framework discussed by assigning an implied volatility for each uncertainty rating and assigning our consistent set of batting averages across all uncertainty ratings. We determined the value of these variables by triangulating among the historical data of our analysts performance in assigning uncertainty, the distribution of returns by uncertainty grouping, the market data of the implied volatility of companies with widely agreed-upon uncertainty characteristics, and a consistently increasing relationship across uncertainty ratings. The 2- and 4-star breakpoints were determined by optimizing the trading performance of a buy-and-sell strategy based on these ratings. We then we rounded the numbers to the nearest 5 percentage points and adjusted them slightly to fit simple, easy-to-remember relationships. 7

8 Quantitatively, the relationship among the implied volatility associated with each uncertainty rating and star rating, the batting average associated with each Consider Buying or Selling rating, and the price/fair value ratio at which we would recommend buying or selling a stock is shown in the following table. Implied Volatility Star Rating Q Percentile of Probability Distribution 75th Implied Volatility 1 Star: 2-5 Star: Low Medium High Very High QQ 58th QQQ Price to Fair Value QQQQ 42nd Ratio (%): >125 >135 >155 >175 QQQQQ 25th <80 <70 <60 <50 Appendix 1: Supporting Data To validate our methodology, we examined our analysts ability to assign uncertainty ratings to the company s value. We took a monthly snapshot of all of the companies under coverage in each uncertainty rating and examined, for each date, the distribution of prices three years in the future, relative to the median price three years in the future. For example, when we grouped low-uncertainty stocks and looked at the distribution of prices three years in the future, the bottom 5th percentile stock traded for 40% of the price of the median stock at the end of the three-year period. Our sample began in August 2001, when we instituted the predecessor of the uncertainty rating, the business risk rating, and the last snapshot was taken in August 2007, which was necessary to examine results over the three-year horizon to August The results are shown in the following table. 8

9 Three Year Stock Price Distribution Relative to Median Return Uncertainty Rating *Percentile Low Medium High Very High Source: Morningstar * Within the group of low uncertainty stocks, the bottom 5th percentile of stock returns was 40% of the median value for all low uncertainty stocks over the three-year time horizon. The results show wider distributions for higher uncertainty ratings, demonstrating that our analysts are capable of grouping companies by uncertainty. In addition, the observed distributions are not symmetric and are well approximated by a distribution that follows the midpoint between a log-symmetric distribution and a Gaussian distribution. K 9