Creating a “Smart” Conditional Consensus Forecast

The study examined analyst forecasts of 26 macroeconomic statistics for August 1998 through March 2007. The four research questions were, (1) Does forecast accuracy persist? (2) What are the determinants of such persistence? (3) Do analysts who exhibit these characteristics make more accurate forecasts than the simple consensus? (4) Is a “smart” consensus that is based on individuals exhibiting these characteristics more accurate than the simple consensus? It was shown that analyst forecast accuracy persists and is determined by long-term past accuracy and the analyst’s overall ability in forecasting all statistics.Forecasts of key macroeconomic statistics are an important input in the decision-making process of security analysts, portfolio managers, and investment advisers. A large body of research has shown that the consensus of macroeconomic forecasts is more accurate than most individual forecasts underlying it, from which comes the strong reliance of the investment community on the consensus. This evidence does not preclude the possibility that some analysts can outperform the consensus. Little evidence, however, indicates that a smart consensus based on forecasts of a highly skilled subset of analysts can consistently beat the simple consensus.We examined analyst forecasts of 26 macroeconomic statistics based on surveys conducted by Bloomberg between August 1998 and March 2007. We posed four research questions: (1) Does individual analyst ability to predict macroeconomic data persist? (2) Do systematic aspects of differential ability exist? In particular, do short-term and long-term track records of forecast performance matter, and if so, which one matters more? Moreover, does an individual’s track record in forecasting the macroeconomic statistic in question (“specific ability”) and other macroeconomic statistics (“general ability”) matter, and if so, which matters more? For our last two research questions, we used the forecast by the simple consensus as our benchmark: (3) Do individuals with certain attributes (e.g., long-term track record for all other macroeconomic statistics) outperform the simple consensus? (4) Does a smart consensus conditional on a simple average of forecasts of individuals possessing certain attributes outperform the simple consensus?To examine these questions, we evaluated and classified each analyst’s forecasting performance according to a four-dimensional metric for each macroeconomic statistic and survey period. We repeated and updated evaluations and classifications after each survey period, so for each statistic and analyst, we obtained a time series of performance. On each performance dimension, we classified analysts as “winners” or not based on their performance relative to all other analysts. The first two dimensions were an analyst’s short-term and long-term performance in forecasting a specific statistic, which we refer to as short-term and long-term specific ability or track record. The other two dimensions were the analyst’s short-term and long-term performance with respect to all the other statistics he or she forecasted, which we refer to as short-term and long-term general ability or track record.We provided the following evidence regarding our four research questions. First, analysts do exhibit persistence in predicting macroeconomic data. Second, although all four dimensions of track record helped explain individual accuracy in holdout periods, long-term track record was more important than short-term track record; track records of the other statistics the individual predicted were more important than track records for the statistic in question; and long-term track records regarding the other statistics that the individual forecasted were the most important of the four dimensions. Third, individuals generally did not outperform a simple unconditional consensus, regardless of which performance dimension one used. Fourth, a “smart” conditional consensus (the simple average of all individual forecasters with superior long-term track records for the other statistics predicted) outperformed both the mean and median consensus forecasts.