This paper examines the natural-rate hypothesis and the Keynesian persistence mechanism using genuine quarterly macroeconomic data for Sri Lanka. Inflation expectations are constructed recursively with an AR(4) forecast using information available at time t. Inference combines classical OLS, HC3 covariance estimates, Newey-West HAC(4) intervals, and a 1,000-replication circular block bootstrap. Preliminary ADF, KPSS, and Zivot-Andrews tests show that unemployment, real GDP, CPI, and real consumption exhibit substantial persistence in levels, whereas year-on-year GDP growth and inflation are more compatible with stationarity. The direct levels model has weak fit (R² = 0.155), pronounced serial correlation, and parameter instability; its positive real-income coefficient is not significant under HAC or block-bootstrap inference. A dynamic model performs materially better (R² = 0.541) and estimates unemployment persistence at 0.705 (HAC 95% CI: 0.508-0.903), while rejecting the unit-persistence restriction ρ = 1 (p=0.0045). GDP growth has the Keynesian/Okun-consistent negative sign but is statistically imprecise. The consumption results do not support persistent fractional long memory: Engle-Granger cointegration is not supported (p=0.793), and the exploratory Hurst estimate for changes in log consumption is 0.395. Overall, the quarterly evidence supports state dependence and medium-run persistence, but not full hysteresis, a stable negative income coefficient in levels, or persistent fractional long memory. The results therefore motivate a more conditional interpretation of Keynesian policy effects in a small open economy.
Keywords: natural rate of unemployment; hysteresis; inflation expectations; Okun relation; quarterly time series; Sri Lanka; robust inference
JEL classification: C22 E24 E31 E32 O53
Download