Measuring the Persistency of Earnings Components: Applications of the VAR Model to Long-Run Japanese Data

Keiichi Kubota, Hitoshi Takehara

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

This study investigates the time-series properties of accounting earnings and their components. We propose a new measure of earnings persistency in accordance with the vector autoregressive (VAR) model–linked earnings and stock returns. As a preliminary analysis, we estimate the first-order autocorrelations and test the stationarity of five variables: earnings, cash flows from operations, total accruals, current accruals, and noncurrent accruals. We then confirm that earnings and noncurrent accruals have a more persistent time-series than cash flows and current accruals. Next, we formulate and estimate the first-order autoregressive model composed of the three variables of utmost interest to accounting researchers, namely, cash flows, current accruals, and noncurrent accruals, and explore how future predictions of these three earnings components are affected by unit impulse shocks. Given the results of the impulse response function analysis, we forecast changes in stock prices based on future innovations of these components, finding that a 1% unit shock in the earnings components affects stock prices by 2% to 2.5%. Finally, we are able to demonstrate excess returns by using the portfolio formation method based on our measure of persistence.

Original languageEnglish
Pages (from-to)329-342
Number of pages14
JournalJournal of Accounting, Auditing and Finance
Volume34
Issue number2
DOIs
Publication statusPublished - 2019 Apr 1

Keywords

  • accounting accruals
  • earnings innovations
  • impulse response functions
  • persistency of earnings
  • vector autoregressive model

ASJC Scopus subject areas

  • Accounting
  • Finance
  • Economics, Econometrics and Finance (miscellaneous)

Fingerprint Dive into the research topics of 'Measuring the Persistency of Earnings Components: Applications of the VAR Model to Long-Run Japanese Data'. Together they form a unique fingerprint.

  • Cite this