Inflation in Mozambique: empirical facts based on persistence, seasonality and breaks

Abstract

This article investigates inflation in Mozambique using long-range dependence (LRD) techniques in monthly data from December 1995 to October 2012. Two important features of the data are analysed: persistence and seasonality, looking at aggregated and disaggregated data. The stability of the parameters across the sample is also investigated. The results indicate a high degree of persistence in the data along with a strong seasonal pattern. Policy implications are discussed.

Keywords:

• Mozambique
• inflation
• long memory

JEL Classification:

• C22

Acknowledgement

Comments from the editor and two anonymous referees are gratefully acknowledged.

Funding

Luis A. Gil-Alana gratefully acknowledges financial support from the Ministry of Economy of Spain [ECO2011-2014 ECON Y FINANZAS, Spain].

Notes

¹ Research with time series in development economics is common; recent studies include Andersson et al. (2013), Erten and Ocampo (2013), Perez-Moreno (2011) and Fields and Puerta (2010).

² There are few previous studies on African macroeconomics; they include Hassanain (2005), Aiolfi et al. (2011), Fiess et al. (2010), Brito and Bystedt (2010), Cermeño et al. (2010), Peltonen et al. (2012) and Coleman (2012).

³ Other recent papers examining long memory and structural breaks in inflation rates are Gadea et al. (2004), Franses et al. (2006), Gil-Alana (2008) and Gil-Alana and Moreno (2012), and forecasting issues are examined in Franses and Ooms (1997) and Barkoulas and Baum (2006).

⁴ Note that the fractional integration framework also admits an infinite moving average (MA) representation as the one presented in Equation 7.

⁵ Performing other standard methods of fractional integration (Sowell, 1992; Beran, 1995), the results were practically the same as those presented here and based on the method of Robinson (1994).

⁶ This is a nonparametric approach of modelling I(0) disturbances that produces autocorrelations decaying exponentially as in the AR case (see Bloomfield, 1973).

⁷ Note that Robinson’s (1994) approach allows us to test any real value of d and thus including nonstationary (d ≥ 0.5) hypotheses. Thus, the analysis is conducted here based on the log-prices series. Similar results were obtained if the estimates were computed on the inflation rate series though in these cases we had to add 1 to get the values corresponding to the log prices. Moreover, in the semiparametric approach performed later, the analysis was conducted on the inflation rates (first differenced data).

⁸ In particular, we used Box–Pierce and Ljung–Box–Pierce statistics (Box and Pierce, 1970; Ljung and Box, 1978) to test for no serial correlation as well as LR tests and other likelihood criteria (AIC, BIC).

⁹ Seasonal dummy variables were also taken into account and the coefficients were found to be statistically insignificant in all cases. Alternatively, we could have used seasonally adjusted data. However, seasonal adjustment procedures have been widely criticized in the time series literature based on the argument that their statistical properties are difficult to assess from a theoretical viewpoint (Ghysels, 1988; Barsky and Miron, 1989; Braun and Evans, 1995).