Long Memory Analysis of Daily Average Temperature Time Series

Abstract

A time series has a long memory, in this case there is autocorrelation at long lags. If a time series display long memory, they show significant autocorrelation between observations widely separated in time. R-software has been used to analyze long memory of daily temperature series of Sokoto metropolis. The Modified Rescaled Range (R/S) statistic, the Periodogram and the Aggregated Variance Methods are used to detect long memory property of the series. Application of these tests suggests that the daily average temperature series shows evidence of long memory.

Keywords: Long memory, Hurst exponent, Aggregated Variance, Modified Rescaled Range and Periodogram.