Inflation uncertainty and economic growth: evidence from the LAD ARCH model

Abstract

In this paper, we combined the panel data and least absolute deviation autoregressive conditional heteroscedastic (ARCH) (L ₁-ARCH) model to infer on the relationship between inflation uncertainty and economic growth in five emerging market economies. Two interesting findings emerged from the analysis; first, we confirmed that the inflation uncertainty has a significant and negative effect on economic growth. Second, inflation is also an important variable and it is detrimental to economic prospects in the fast-growing Association of Southeast Asian Nations (ASEAN) economies. All in all, the empirical findings suggest that greater stability in the economy may be desirable in order to stimulate economic growth in the region.

Keywords:

• inflation uncertainty
• economic growth
• LAD ARCH

Acknowledgements

A shorter version of this paper was presented at the Asian Development Bank (ADB) meeting in Bangkok. The authors acknowledged the helpful comments on an earlier version of the paper from anonymous referees. We are also grateful to A.H.M. Rahmatullah Imon, Sung K. Ahn and Stilainous Fountas for their comments and suggestions on an earlier draft. Financial support from the ADB and the Malaysian Government are gratefully acknowledged [Grant no: 06/01/07/0241RU]. The usual caveats apply.

Notes

A survey of the empirical work on the growth uncertainty nexus by Holland 16 revealed mixed results. The ambiguous results are due to factors like the statistical method used to measure inflation (nominal) uncertainty, the chosen econometric methodology, the country and the sample period.

After the devaluation of the Thai baht (July 1997), other ASEAN countries abandoned their close links with the US dollar and began to depreciate. The most severe pressures in foreign exchange markets, in the third quarter of 1997, were experienced by Thailand, Malaysia, Indonesia and the Philippines, but the currencies of Singapore and a number of Asian countries also weakened. Apart from the sharp deterioration in exchange rates, most of these countries experienced higher interest and inflation rates in the post-crisis period. For example, inflation rates in Indonesia and Thailand in the first 12 months of the crisis were 56.6 and 10.8, respectively. Inflation rate for the other Asian countries (namely, Malaysia, Philippines and South Korea) were also higher but recorded single-digit inflations (inflation rate of about 9.9% to 6.3%). These observations are viewed as outliers and clearly coefficient estimates based on standard models are affected by these few observations.

To measure the level of uncertainty, some researchers have relied on survey data. The dispersion of individual forecasts is used as the measure of inflation uncertainty; see Holland 16 for articles that used this technique. The major problem with this measure is that it is unable to distinguish between variability and uncertainty. In other words, it includes both the predictable and unpredictable variability (uncertainty) and as such it has little to do with the fluctuation in uncertainty.

For recent application of the GARCH model to estimate inflation uncertainty and the impact of inflation uncertainty on output growth, see Fountas and Bredin 10. Although the model is commonly used in empirical studies, some authors have pointed out that the ARCH-based measures can yield misleading account to the changes in uncertainty and the effect on macro variables.

Least absolute deviation (LAD) estimates are more efficient than the LS estimates for models with errors having heavy-tailed distribution 28. We are grateful to an anonymous referee for pointing this out.

See Jiang et al. 18 for a more detailed discussion on the properties of robust ARCH models.

The authors gratefully acknowledge Fountas for providing the data to construct the volatility series. The data spanning over 50 years covers a variety of inflation experiences in the region. Existing literature has pointed out that those outliers (regimes changes) will occur with higher probability for longer sample periods. Thus, the robust methods offer advantage over the standard GARCH specification in the assessment of the hypothesis.

We note that reliable GDP data prior to 1980 are unavailable for some of these countries under investigation.

For further discussion on analysis of panel data, see Hsiao 17 and Judson and Orphanides 20.

The measure of volatility of inflation involves ARCH modeling. To employ the L ₁-ARCH model, we first need to calculate the residual of the inflation rate, r _ij =π _ij -median (π), where π _ij is the inflation series in country i and year j.

We discovered that inflation rate appears as I(0) process in all the ASEAN countries. This means that the effect of any shocks on the series is only transitory and as such making the need for policy action less mandatory.

To conserve space, the results are not reported here since the model did not fit the data. They are available from the authors upon request.

The difficulties in employing some form of GARCH modeling strategy are recognized in the recent literature. It is worth noting that Kramer and Azamo 21 and others argue that changes in the variance could arise from changes in the mean. They demonstrate that the estimated persistence parameter in the GARCH (1,1) model contains upward bias when the researcher ignores structural break in the mean. We do not pursue the idea in this paper.

The article by Higgins and Bera 15 relied on nonlinear ARCH (NARCH) model to deal with the thick-tailed error terms. They pointed out that the NARCH limits the influence of large residuals essentially in the same way as the estimator based on robust methods.

To generate the bootstrap standard error, we relied on the procedure as outlined in De Angelis et al. 5. Briefly, we let the initial bootstrap observations be, and the initial bootstrap errors, , as in Equation (1). The bootstrap estimates and of αˆ and βˆ are then computed from Equations (2) and (3) with y _t replaced by , as in Equation (1) for t=1, …, n, and replaced by , where are bootstrap samples of u _t (random samples). Repeat this process B times, yielding and and the subsequent bootstrap standard errors for the parameters are given by and , and are the respective means of and . Here, B=1000.

The specification before this assumes that the error term is uncorrelated with the regressor y _ij . Some authors, however, have argued that an increase in output above its full employment (natural) level would result in inflationary pressure (also known as the Phillips curve effect). This means that the direction of caution runs from higher growth to higher inflation (reverse causality). To address this issue of simultaneity or reverse causality, we relied on instrumental estimation method. We thank an anonymous referee for proposing this analysis to us.