to some benchmark portfolio. The risk of a managed portfolio relative to its benchmark is calculated by computing the standard deviation of the difference between the managed portfolio and benchmark returns. This yields the relative risk measure that is commonly referred to as tracking error. Suppose that a manager currently uses the market model to measure risk and attribute return. The manager's stock selection ability can be measured by the residual return-the difference between the managed portfolio return and the market's systematic component (market beta times market return). The standard deviation of the systematic component is a measure of systematic risk. The standard deviation of the residual return measures residual risk. Residual risk can be further decomposed into residual factor (i.e., the factors that can explain the residual returns) and residual specific risk. Figure 20.5 presents a flowchart of the risk estimation process supporting a U.S. equity factor model. This process, which is presented in eight parts, extends to other markets as well as across markets. We explain each of the eight steps in the risk estimation process: Step 1 Source and collect exposure and market data. Exposure data may include industry classifications, fundamental data (e.g., book values), earnings estimates, and macroeconomic variables. Market data refers to daily asset prices and total returns as well as trading volume. Step 2 Transform the raw exposure information and market data into asset exposures that will be used to estimate the parameters of the factor model. In the U.S. market, there are three basic types of exposures: (1) industry, (2) investment style, and (3) market. Step 3 Construct an exposure matrix21 (for each factor we have one exposure per asset) that consists of industry exposures, style exposures, and a possible new set of exposures that are generated from running time-series regressions of stock returns on market returns. This regression generates a market beta that we can use to measure a stock's exposure to the local market. Step 4 Determine which assets qualify for our estimation universe. The estimation universe represents a set of assets that will be used to determine the parameters of the factor model (i.e., the factor returns). The estimation universe may change quite a bit over time. One reason why an asset would not qualify for a particular estimation universe is that there is not enough historical information on its stock returns and/or company information. Step 5 Estimate factor returns by running a cross-sectional regression of asset returns-as defined by the estimation universe-on their factor exposures. The coefficients in this regression are estimates of factor returns. Step 6 Run cross-sectional regressions each day over some sample period. Repeating these regressions over successive periods of time generates a time series of factor and specific returns. We use these time series of returns to estimate 21If we had N stocks and K factors then our exposure matrix would be an N x K matrix of exposures.