In this section we report on a series of sensitivity analyses of our benchmark model, Equation (3.1). We assess the performance of our estimation in two key dimensions; the data choices of distinct measures, and the model speci-fications. Our qualitative findings are robust to all these sensitivity checks.
Corresponding results are here summarised, with detailed results shown in the appendix.
3.4.1 Different measures of macroeconomic indicators
While focusing on the specific model variations and re-estimations, the first set of robustness checks is concerned with different measures of the macroe-conomic indicators used in the benchmark estimation.
Measures of the output-gap
The first variation is related to possible differences in the measurements of the output-gap due to different filtering techniques. We use the standard Hodrick-Prescott (HP) filter in our benchmark model. As a robustness test, we obtain output-gap measures from a two-sided band-pass filter, which also limits the sample size compared to the benchmark model. There are only marginal quantitative differences between the results from the re-estimated benchmark model with band-pass filtered output-gap and the HP-filtered output-gap (see Figures 3.19 to 3.25). Qualitatively, the results of coeffi-cients and contributions are the same. Quantitative differences appear in the coefficients of output-gap that are slightly higher during the GFC across all countries. Furthermore, the median contribution of time variation of output-gap and import price inflation is marginally lower.
Indicators of inflation expectations
The second variation is concerned with possible differences between the es-timates of the long-term inflation trend-expectations and more traditional survey-based expectations. Since survey-based expectations might be sys-tematically biased, we use long-run trend expectations estimates, incorporat-ing information from survey-based expectations. Our results indicate that inflation expectations are a crucial driver for inflation dynamics that can potentially outweigh exogenous cost-push shocks. Therefore, it is impor-tant to verify whether the coefficients and contributions are robust across different measures of inflation expectations. We hence substitute the
trend-expectation estimates with Consensus long-term inflation trend-expectations. Re-estimating our benchmark model using Consensus long-run inflation expec-tations yield qualitatively equal results of coefficients and contributions (see Figures 3.26 to 3.32). As revealed in Figure 3.4, trend expectations and con-sensus expectations display some quantitative differences, depending on the point in time and the respective country. Overall, Consensus survey expec-tations are systematically higher than the long-run inflation trend estimates.
This is reflected in the median contributions of the forward-looking com-ponent using Consensus expectations compared to the benchmark results.
The median contributions related to the time variation in parameters are quantitatively slightly more pronounced for the output-gap and import price inflation, and less pronounced for the forward-looking component. Estimated coefficients, however, depict almost no quantitative differences compared to the benchmark results.11
Import price measures
Import price inflation data might depend on whether it is retrieved from terms of trade or obtained from national accounting. Even though data quality and provision has been improving constantly, and depending on the country, there are still substantial differences across the import price series retrieved from different sources. Thus, we also re-estimate our benchmark model with import prices obtain from the WEO database as a third variation.
Overall, the qualitative implications of the re-estimated coefficients and
con-11Results of Singapore exactly match the benchmark results, as we could not use trend inflation estimates for Singapore in the benchmark estimation.
tributions (see Figures 3.33 to 3.39) are in line with our benchmark model es-timates and respective contributions. The coefficients of the forward-looking component are systematically higher, and hence the correlation between the DE transparency index and the coefficients on the forward-looking compo-nent is altered to 0.69. Although the combined contribution of oil and import price inflation is quantitatively similar to the benchmark results, the weight between the contributions of import and oil price inflations shifts towards the latter. Resultantly, the oil price inflation constitutes a much larger share of the median - and also country-specific - contribution than import price inflation.
3.4.2 Model specifications
The second set of robustness checks is concerned with the model specification.
Coefficients of non-oil-import and oil price inflation
The fourth variation is related to possible differences of coefficients of non-oil-import price inflation and oil price inflation in the Phillips curve. As mentioned in the estimation section, we only estimate three parameters for our benchmark model: a coefficient for the forward-looking dynamics, a coef-ficient for the economic slack that represents the slope of the PC, and one co-efficient for overall import price inflation, averaging across different dynamics of non-oil-import prices and oil prices. We decompose the contributions us-ing the import price coefficient for both series ex-post. The reasonus-ing for our benchmark procedures is that import price inflation should in principle con-tain oil price inflation for oil importing countries. Since not all the ASEAN-5
countries are oil importers though, we explicitly include non-oil-import price inflation and oil price inflation in the model for the fourth variation, estimat-ing the four parameters.
Results from the model with four parameters (see Figures 3.40 to 3.46) depict much higher contributions of the residuals to the median as well as country-specific headline inflation, whereby median contributions account for near 30%, compared to roughly 10% in the benchmark estimation. Moreover, the contribution of oil price inflation is substantially altered, whereby the contri-butions of forward-looking dynamics are quantitatively lower in comparison with the benchmark results.
The country-specific coefficients of non-oil-imports are systematically higher (by roughly 0.05), but they reveal the same dynamics as in the benchmark case. For the oil-exporting countries (Singapore and Malaysia), the coeffi-cients of oil price inflation are significantly positive throughout the sample period, but reveal little time variation. The oil price inflation coefficients are small and do not change significantly over time across the countries -the significant levels of coefficients range from 0.005 to 0.028. The remain-ing coefficients are quantitatively negligible in difference. Overall, the fourth specification reveals qualitatively similar results to the benchmark estima-tion.
Model including oil price inflation and exchange rate
As most of the ASEAN-5 economies are highly open economies, exchange rate movements might have a relevant pass-through to headline inflation.
In the last robustness check, we augment the model specification with the exchange rate in addition to expectations, economic slack and oil price infla-tion. In line with the empirical work done by Devereux and Yetman (2014), exchange rate pass-through is very limited for the ASEAN-5 countries (see Figures 3.47 to 3.53). The coefficient on exchange rates is significant across countries and time.12 The median contributions of the forward-looking com-ponent only reveal slight quantitative differences compared to the benchmark model. The median and country-specific contributions of the exchange rate are quantitatively very small. The median and country-specific contributions of oil price inflation, however, are larger than in the benchmark model.