Multi-step-ahead quarterly cash-flow prediction models

Kenneth S. Lorek, G. Lee Willinger

Research output: Contribution to journalArticlepeer-review

11 Scopus citations


We provide new empirical evidence supportive of the Brown-Rozeff ARIMA model as a candidate univariate statistically based expectation model for multiperiod-ahead projections of quarterly cash flows. It provides 1- through 20-step-ahead projections of quarterly cash flows that are significantly more accurate than those generated by the premier multivariate quarterly time-series, disaggregated-accrual regression model popularized by Lorek and Willinger (1996). We also find that both quarterly earnings and quarterly cash flow from operations are modeled by the same Brown-Rozeff ARIMA structure, although the autoregressive and seasonal moving-average parameters of the quarterly earnings model are significantly larger than those of the cash-flow prediction model. This finding is consistent with Beaver (1970) and Dechow and Dichev (2002), among others, who argue that accounting accruals induce incremental amounts of serial correlation in the quarterly earnings time series vis-à-vis the time series of quarterly cash flows. Such findings may be of interest to analysts who wish to derive multi-step-ahead cash-flow predictions, and accounting researchers attempting to adopt a statistical proxy for the market's expectation of quarterly cash flows. Finally, we propose a forecasting schema by which statistically based cash-flow forecasts are adjusted upwards or downwards via qualitative assessments regarding the economy, industry, and firm by analysts employing fundamental financial analysis.

Original languageEnglish (US)
Pages (from-to)71-86
Number of pages16
JournalAccounting Horizons
Issue number1
StatePublished - Mar 2011


  • Arima models
  • Cash-flow forecasts
  • Multivariate time-series regression models
  • SFAS No. 95

ASJC Scopus subject areas

  • Accounting


Dive into the research topics of 'Multi-step-ahead quarterly cash-flow prediction models'. Together they form a unique fingerprint.

Cite this