ARIMA earnings forecasts versus the announcement-window price reaction

Three questions at the print

Editorial framing for this archive piece: an earnings release is easier to read if three questions stay apart. Did a time-series model beat the consensus estimate on the earnings-per-share print. Did the rest of the fundamental overlay agree with that print. Did event-driven context on the prevailing tape leave room for the earnings surprise to appear in price.

The historical workflow treated those as separate measurements. A closer forecast residual was not carried forward as evidence about announcement-window price direction.

How the announcement window was measured

The announcement window is the short interval from the session before an earnings release to the session after it. The historical study measured the announcement-window price change as the difference between the last price before the earnings date and the first price after it, because many results arrive before the cash session opens.

A positive earnings surprise versus the consensus estimate was not, in that sample, a sufficient condition for a higher post-announcement price. An earnings surprise is the gap between reported earnings-per-share and a previously published estimate. The consensus estimate is the published average of sell-side earnings or revenue forecasts used as the public benchmark at the announcement.

The ARIMA earnings path

The independent earnings path was an ARIMA model. ARIMA is a time-series specification that forecasts a series from its own lagged values, lagged forecast errors, and differencing that removes drift or trend. Autoregressive terms, moving-average shock terms, and differencing were chosen to handle lag structure and drift in the quarterly series.

The selected specification was the one with the smallest gap between forecasted and realized earnings-per-share over the most recent four quarters when projecting the next quarter.

Forecast residual versus the consensus estimate

The forecast residual is the signed difference between a model forecast of earnings-per-share and the number later reported. On the same historical earnings series, the consensus estimate underestimated actual earnings-per-share by 3.8 cents on average, while the ARIMA forecast underestimated it by 1.01 cents.

Even when the ARIMA forecast was closer to realized earnings-per-share than the consensus estimate, that gap was not treated as evidence that the announcement-window price change would be positive.

Fundamental overlay beside the print

A fundamental overlay is additional reported items, such as revenue versus estimate and forward guidance, that sit beside the earnings-per-share print. Reported revenue versus consensus and management guidance were treated as additional fundamental overlays that the market can weight alongside the earnings-per-share print.

Among more than 1,400 companies reporting in April 2014, 58 percent exceeded the consensus earnings-per-share figure and 54 percent exceeded the consensus revenue figure, while the contemporaneous market reaction was still described as negative.

When the tape dominates the name

Event-driven context is the market-wide tape and release-day setting that can dominate a single-name reaction to a scheduled announcement. Post-announcement price direction was described as depending on the prevailing market regime. A declining tape made a positive surprise more likely to coincide with a decline, and a rising tape the reverse.