An empirical investigation of Bangladesh's inflation dynamics: evaluating persistence and identifying structural breaks

The study investigates Bangladesh’s inflation dynamics given concerns that inflation and its uncertainty can impair resource allocation, particularly in developing economies. It motivates the analysis by recent global inflation surges linked to supply chain disruptions and policy responses, emphasizing the importance of persistence and stationarity for monetary policy design. The paper focuses on three research questions: (1) identifying structural breaks in Bangladesh’s inflation and their causes; (2) measuring inflation persistence across sub-samples defined by breaks and interpreting policy implications; and (3) examining causal links between inflation and inflation uncertainty in time-invariant and time-varying settings. The work is positioned to inform policymakers on forecasting, policy transmission, and expectation anchoring in an economy increasingly exposed to global shocks.

The literature distinguishes forms of inflation persistence, with policy-relevant definitions tied to the speed of convergence following shocks. Persistence shapes the sacrifice ratio and policy space, and reduced-form persistence may differ from structural persistence stemming from policy regimes or macro drivers. Ignoring structural breaks biases persistence upward and unit root tests toward non-rejection (Perron, 1990), making break detection critical (Bai-Perron). Empirical work often finds lower persistence when breaks or regime shifts are accounted for; inflation targeting tends to reduce persistence in several economies. Measures include AR coefficients and spectral density at zero frequency, with AR(1) commonly used; long-memory approaches like ARFIMA also appear in prior studies. Time-varying methods (rolling regressions, TVP-VAR, bootstrap break tests) reveal changing persistence associated with policy changes and crises. The inflation–uncertainty nexus presents mixed evidence: Friedman-Ball posits inflation raises uncertainty; Cukierman-Meltzer posits uncertainty raises inflation; other studies find negative or U-shaped relationships or regime-dependent effects. Recent work using breaks and time variation shows periods where inflation causes uncertainty (e.g., Turkey) and varying signs across stable vs. crisis periods. For emerging markets, evidence often supports a positive link and at times bidirectional causality.

Data comprise monthly CPI inflation for Bangladesh from June 1997 to July 2021 (IMF International Financial Statistics; T=290). Descriptive statistics show mean inflation of 6.25% (range 1.12%–17.15%) and significant autocorrelation in early lags (Ljung-Box Q). Unit root properties were assessed via ADF, Phillips–Perron, DF-GLS, and the Kapetanios–Shin–Snell (KSS) test accommodating multiple breaks; all reject a unit root at 1% significance. Structural breaks in the inflation mean were detected using Bai-Perron (1998, 2003) multiple break tests with trimming ε=0.15 and up to five breaks, employing SUP-F, UDmax/WDmax, and sequential tests; information criteria (BIC, LWZ) guided model selection. Sequential testing indicated three breaks (June 2003, March 2007, April 2012). Persistence was measured by: (i) AR(1) with drift, estimated on the full sample and sub-samples split at break dates; and (ii) ARFIMA to allow fractional integration (d), interpreting d per standard rules (stationary mean-reverting when 0≤d≤0.5). Inflation uncertainty was modeled using ARMA(p,q)-GARCH(1,1) where lag orders were chosen by BIC. Structural break dummies were included in the variance equation to account for regime shifts; the conditional variance h_t from GARCH served as the uncertainty series. Granger causality (time-invariant) tested directional predictability between inflation and uncertainty at various lags (0, 2, 4, 8). Time-varying Granger causality followed Shi et al. (2018) using rolling and recursive windows with Max-Wald statistics to detect evolving causal links.

• Stationarity and dependence: ADF, PP, DF-GLS, and KSS reject a unit root at 1%; Ljung-Box Q indicates significant autocorrelation in early lags, consistent with persistence; heteroskedasticity supports GARCH modeling.
• Structural breaks: Multiple-break testing (Bai-Perron) finds evidence for up to five potential breaks; sequential testing selects three: June 2003, March 2007, and April 2012.
• Persistence levels: AR(1) coefficient for the full sample is high (≈0.90). Persistence declines in two sub-samples: 2003m7–2007m4 (β≈0.59) and 2007m5–2012m5 (β≈0.74). Post-2012 (2012m6–2021m7), persistence rises markedly (AR(1)≈0.92). ARFIMA estimates indicate stationarity with long memory not dominant (d≈0.49 overall; declines in sub-samples where AR(1) falls), consistent with mean reversion but high persistence.
• Regional comparison: Bangladesh’s persistence is comparable to, and sometimes lower than, selected South and Southeast Asian peers; ARFIMA estimates are generally lower than AR(1) across countries, implying limited long memory.
• Inflation uncertainty modeling: Without break dummies, ARCH+GARCH sums exceeded one (explosive); including break dummies yielded sums below one, consistent with a stationary conditional variance and credible uncertainty series.
• Causality: Standard Granger tests show inflation → uncertainty at 0, 2, and 4 lags; at 8 lags, bidirectional causality appears. Time-varying Granger tests (Max-Wald) indicate bidirectional causality over time at the 1% level, validating both the Friedman-Ball and Cukierman-Meltzer hypotheses in Bangladesh.
• Descriptives: Mean inflation 6.25% with range 1.12%–17.15% over 1997m6–2021m7.

The three structural breaks align with major economic events and policy shifts: (1) June 2003 is linked to the transition to a floating exchange rate (managed as a dirty float), floods, rising credit demand, and higher global oil prices; (2) March 2007 coincides with the global food price surge, domestic demand-pull pressures during rapid growth, exchange rate pass-through in a net food-importing economy, and supply-chain disruptions from anti-corruption measures; (3) April 2012 marks a downward inflation shift attributable to bumper Boro rice and other crop harvests and later reinforced by the 2014–2016 oil price collapse. Persistence declines during 2003–2007 and 2007–2012, suggesting policy responses and favorable supply developments temporarily reduced persistence. After 2012, persistence increases again, implying both demand- and supply-side forces and heightened exposure to global shocks. Bidirectional, time-varying causality between inflation and inflation uncertainty indicates a reinforcing cycle: higher inflation raises uncertainty, deterring investment and consumption, while greater uncertainty feeds back into inflation through precautionary pricing and wage-setting. For an emerging economy like Bangladesh, this cycle can exacerbate inefficiencies and complicate policy, underscoring the need for expectation management, credible frameworks, and tools to hedge inflation risk.

The paper contributes by jointly analyzing structural breaks, persistence, and the inflation–uncertainty nexus for Bangladesh (1997m6–2021m7). It identifies three breaks (2003m6, 2007m3, 2012m4), documents high overall inflation persistence with notable declines around two breaks, and demonstrates bidirectional, time-varying causality between inflation and inflation uncertainty—supporting both the Friedman-Ball and Cukierman-Meltzer hypotheses. Policy recommendations include proactive and context-specific interest rate management; adopting a Medium-Term Fiscal Framework to improve fiscal-monetary coordination and reduce persistence; enhancing central bank communication and inflation expectations surveys (including items on persistence, uncertainty, energy/food, exchange rates, and partner-country inflation); fostering inflation-hedging financial instruments; and considering a flexible monetary policy framework with broad inflation objectives tailored to domestic conditions rather than strict inflation targeting. These insights generalize to other developing economies facing similar persistence and uncertainty challenges and suggest future work decomposing inflation into demand- and supply-side components for more granular policy design.

The analysis is constrained by limited granularity of Bangladesh’s inflation data, precluding decomposition into detailed supply- and demand-side components and restricting sectoral analysis. Cross-country comparisons of persistence are indicative rather than strictly comparable due to differences in CPI baskets, weights, and methodologies. Despite using break-adjusted methods, break detection and model selection can be sensitive to trimming choices, small-sample properties, and information criteria, potentially affecting estimated break dates and persistence levels.