Abstract
This paper provides a thorough analysis on the dynamic structures and predictability of China's Consumer Price Index (CPI-CN), with a comparison to those of the United States. Despite the differences in the two leading economies, both series can be well modeled by a class of Seasonal Autoregressive Integrated Moving Average Model with Covariates (S-ARIMAX). The CPI-CN series possess regular patterns of dynamics with stable annual cycles and strong Spring Festival effects, with fitting and forecasting errors largely comparable to their US counterparts. Finally, for the CPI-CN, the diffusion index (DI) approach offers improved predictions than the S-ARIMAX models.
Introduction
Predicting inflation is a cornerstone of monetary policy, but modeling China's Consumer Price Index (CPI-CN) presents unique hurdles. Unlike the United States, the official Chinese CPI is not seasonally adjusted at the source. This lack of adjustment is problematic because China’s most significant economic "shock"—the Spring Festival (Lunar New Year)—follows a lunar calendar. Because the holiday shifts between January and February each year, standard solar-calendar seasonal models often fail to capture the resulting price volatility. Furthermore, historical skepticism regarding the quality of Chinese macroeconomic data necessitates a rigorous, comparative framework to determine if CPI-CN follows regular dynamic structures similar to the US and if its predictability holds up under scrutiny.
Data Composition and Trends
The study utilizes data spanning from January 2002 to November 2016. A fundamental difference between the two economies lies in the weighting of CPI categories. As illustrated in Figure 1, China's CPI is heavily dominated by food (31.21%), whereas the US CPI is dominated by housing (42.09%). This structural difference means that Chinese consumer prices are inherently more sensitive to agricultural supply and upstream production costs.
Despite these structural differences, both indices show consistent upward trends over the study period. Figure 2 illustrates the raw CPI series and their respective seasonal components. The seasonality in CPI-CN is notably more pronounced and regular than that of the US, largely due to the impact of the "wandering" Lunar New Year and the absence of official seasonal smoothing.
Methodology
The researchers employ two primary modeling strategies to evaluate predictability:
- S-ARIMAX (Seasonal Autoregressive Integrated Moving Average with Covariates): This standard linear time-series model is augmented with exogenous variables (X) to handle China's specificities. The Spring Festival is partitioned into three sub-periods—before, during, and after—to capture the full cycle of price fluctuations. Additionally, dummy variables account for major one-off outliers, such as the 2008 Southern China snow disaster.
- Diffusion Index (DI) Approach: This method uses Principal Component Analysis (PCA) to extract common "factors" from a large set of macroeconomic variables (31 for China, including money supply and trade data; and the FRED-MD database for the US). These factors serve as predictors in a regression model to forecast future inflation.
Key Insights and Results
The research yields several critical findings regarding the nature of inflation in these two global powers:
- Holiday Modeling is Essential: Accounting for the Spring Festival as a shifting window rather than a fixed month is vital. This adjustment, along with accounting for the 2008 snowstorm, reduced statistical outliers in the Chinese models from 474 to zero in the top-performing configurations.
- Predictability Parity: Despite different economic drivers, forecast errors for both countries are remarkably comparable. At the 1-month horizon, the forecasting error was 0.51 for China and 0.49 for the US, suggesting that Chinese CPI data is of high quality and reacts predictably to economic events.
- DI Performance Advantage: The Diffusion Index approach provided a clear edge in China for short-term forecasts (1–3 months), reducing Root Mean Squared Error (RMSE) by 10-14% compared to S-ARIMAX. Interestingly, the DI approach performed worse for the US during this period, likely due to instabilities in factor models caused by the 2008 financial crisis.
- Supply-Side Sensitivity: The primary factor in the Chinese DI model was heavily weighted toward the Producer Price Index (PPI) and food items, confirming that Chinese consumer prices are highly sensitive to upstream production costs.
Conclusion
China's CPI is just as predictable as that of the United States when local specificities, particularly the Lunar New Year, are properly modeled. While structural differences exist—notably the higher sensitivity of CPI-CN to food and production costs—the underlying dynamic structures are robust. The DI approach offers a superior tool for short-term inflation forecasting in the Chinese context, outperforming standard univariate models.
Summary updated