Forecasting
A core function in firms is the ability to forecast.
An investment firm wants to forecast the prices of future investments and the probability of future events that may affect prices. Government and health services may wish to forecast the spread of a disease. A university will want to forecast student numbers.
There are a range of problems that may derail our ability to forecast - many that we have seen during this subject - including errors due to heuristics, overconfidence (particularly overprecision and overestimation), and noise.
Forecasting also suffers from the problem that knowledge is distributed across the firm. For instance, sales team members may get detailed knowledge of what customers think about the new product. How can you effectively gather this information for use in the forecast?
The following chapters explore various approaches to improving forecasting accuracy, with a focus on methods that aggregate distributed knowledge:
In Expert Political Judgment, we examine Philip Tetlock’s research on expert forecasting performance, including the distinction between “foxes” (who draw on diverse information sources) and “hedgehogs” (who rely on a single big idea).
Words of estimative probability addresses the ambiguity in verbal probability expressions and demonstrates why numerical probabilities are essential for precise forecasting.
The outside view introduces reference class forecasting as a powerful technique to overcome the planning fallacy and other biases by basing estimates on similar past cases rather than detailed project specifics.
Better forecasting explores techniques to improve forecast accuracy, primarily through evidence from the Good Judgment Project and related research, including training methods for superforecasters.
Prediction markets examines how market mechanisms can aggregate dispersed information to create accurate forecasts through prices, with evidence from corporate implementations and discussion of their limitations.
Forecasting platforms covers non-market approaches like Metaculus and Good Judgment Open, which use statistical methods rather than price mechanisms to aggregate probabilistic judgments.
By the end of these chapters, you will be able to answer the following questions:
- How can forecasting fail?
- What methods might improve our forecasts?
- How can organisations effectively aggregate distributed knowledge for better predictions?
- What are the relative strengths and limitations of different forecasting approaches?
- How can cognitive biases be mitigated in forecasting processes?