Analysis · Finance
When an Earnings Forecast Loses to Assuming No Change
By AWEI · AI-compiled · Published · Analysis prepared · 1 source · doi.org
Keeping future earnings equal to current earnings sounds like a low bar for a forecasting model. Yet in a 2014 study, Kevin K. Li and Partha Mohanram found that this simple baseline beat the established Hou–van Dijk–Zhang model, known as HVZ, on average absolute forecast error across one-, two- and three-year horizons.
The study evaluated forecasting methods using historical company accounts, rather than running an experiment. Its accuracy comparison covered 119,653 firm-year observations from Compustat for 1969–2008. The researchers fitted models using appropriately lagged ten-year windows and allowed for reporting delays, keeping future information out of the forecasts being tested.
They compared HVZ with the unchanged-earnings baseline and two alternatives. The earnings persistence model allowed profits and losses to persist differently. The residual income model added book value and accruals. Such forecasts feed into implied cost of capital: a discount rate inferred by matching a valuation model to the observed stock price.
The residual income model delivered average absolute errors about 28% lower than HVZ at one year and 38% lower at three years. These were reductions in errors scaled by market value or share price, not percentage errors relative to earnings. Comparability mattered: when HVZ was also estimated per share, the reported advantage narrowed to 6%–15% across those horizons.
There were limits to the win. Median errors for the residual income model and the unchanged-earnings baseline were not statistically distinguishable. The accuracy sample required observed future earnings, and the main earnings measure excluded special and extraordinary items. Associations between implied cost of capital and subsequent returns did not establish causation or guarantee future performance.
The practical lesson is to ask what a forecast beats, for which companies, and under which error measure. An unchanged-earnings baseline can expose weak forecasting performance, even though its lack of growth makes it unsuitable for some implied-cost-of-capital calculations. Beating that baseline is useful evidence; it does not validate every downstream use.
Sources used for this article (1)
Direct links to the publisher reports used to prepare this article.
- Source 1
- Evaluating cross-sectional forecasting models for implied cost of capital Kevin K. Li, Partha S. Mohanram
