identify instances of performance or achieved tracking error that are outside of normal expectations. The green zone concept embodies the following elements: 1.    For the prior week, month, and rolling 12 months, we calculate the portfolio's normalized returns, which are defined as excess returns over the period minus budgeted excess returns over such period, all divided by target tracking error scaled for time.22 This statistic might be viewed as a test of the null hypothesis that the achieved levels of excess returns are statistically different from the targeted/budgeted excess returns. 2.    For the prior 20- and 60-day periods, we calculate the ratio of annualized tracking error to targeted tracking error. In this test, we examine whether the variability in excess returns is statistically comparable to what was expected.23 Note that there is no one correct period of time over which to measure tracking error. While for the purposes of this chapter we have selected a shorter-term horizon, strong arguments can be made for including longer-term horizons as well. The point here is that unusual blips in volatility may serve as filters for identifying anomalous environments in which underlying risk dimensions may be undergoing profound change. This tool is designed to help management and portfolio managers ask better and timelier questions. As an example of this point, consider Figure 17.5, which shows the time series of predicted tracking errors juxtaposed against rolling 20- and 60-day tracking errors. Not surprisingly, the 20-day measure is more volatile than the 60-day measure and is therefore more responsive to changes in market behavior. The challenge for the risk monitoring professional is to ascertain whether the signal is anomalous or whether it carries information content that should be acted upon. At GSAM, we use this signal as a basis for initiating dialogue between the RMU and portfolio managers to better understand the causes behind these two signals and their consequences. 3. For each of the calculations in fl) and (2) above, we form policy decisions about what type of deviation from expectation is large enough, from a statisti cal standpoint, to say that it does not fall in the zone of reasonable expecta tions that we call the green zone. If an event is unusual, but still is expected to occur with some regularity, we term it a yellow zone event. Finally, red zone events are defined as truly unusual and requiring immediate follow-up. The de finition of when one zone ends and a second begins is a policy consideration that is a function of how certain we would like to be that all truly unusual events are detected in a timely fashion. For example, if the cost of an unusual 21Refer to an article entitled: "The Green Zone . . . Assessing the Quality of Returns," by Robert Litterman, Jacques Longerstaey, Jacob Rosengarten, and Kurt Winkelmann of Goldman Sachs & Co. (March 2000). 22For example, in calculating the monthly normalized return, the denominator consists of the annual tracking error target divided by the square root of 12. 23This test is analogous to ANOVA techniques (e.g., the "F" test) in which one looks at the ratio of variances to determine whether they are statistically comparable. In this case, we are examining the ratio of standard deviations.