Base-rate neglect is one of the classical cognitive biases affecting expert estimates. We have identified base-rate neglect as an important bias in the interest-rate forecasts of a small regional bank. Experts there produce forecasts for the returns of German government bonds with maturities of 3 months and 2, 5, and 10 years, respectively, for up to five years into the future. Their main input are economic studies of major banks. Almost all of the forecasts compare badly with the ex-post realizations of the bond returns.
An interview-based investigation of the forecast process revealed that none of the experts included historic time series into the forecast. These were readily available in the bank, and showed clear trends opposite in direction to the direction of the forecasts. Subsequently, a minimally invasive correction scheme was set up which was mainly based on visualization, accounting for the limited statistical capacities of the bank: (i) perform the forecast as usual, (ii) receive information on trends derived from time series, (iii) estimate a (credibility) weight factor for one’s own forecast, and (iv) generate a weighted average of (i) and (ii), thus circumventing the necessity to use Bayesian statistics.
To present the methodology as simply as possible, a picture book was conceived. For facilitating the actual data treatment [step (iv)], the bank’s experts were handed over a painting book whose final drawings later were re-digitized.