Exploring the Chinese Yuan Exchange Rate

Based on monthly data since 2010, we have developed a RMB exchange rate analysis framework that follows time series constraints. Empirical results show that, under a macro monthly information set, the influence of macro variables on the RMB exchange rate mainly manifests in central constraints and interval shaping, rather than in precisely depicting short-term fluctuation paths.

1. Model approach and technical framework. We believe that in the case of exchange rates—typical high-noise, strongly nonlinear problems—the value of models is less about finely fitting historical volatility and more about maintaining certain stability in out-of-sample environments. Since RMB exchange rate fluctuations are often driven by synchronized internal and external factors, and their transmission paths are continually reshaped by macroeconomic changes, a single, stable linear structure is unlikely to remain effective across different periods. Based on this understanding, we introduce a random forest model to analyze USD to RMB exchange rate changes, focusing on the model’s robustness under time series constraints. As a non-parametric ensemble learning method, random forests do not rely on preset functional forms and can capture underlying nonlinear relationships and interactions among multiple variables.

2. Macro variable constraints rather than volatility. Backtesting results clearly indicate that macro variables mainly influence the RMB exchange rate through central constraints, rather than providing precise descriptions of short-term fluctuation paths. Time series segmentation backtests show that the amplitude of the model’s predicted sequences is significantly lower than actual exchange rate volatility. This phenomenon is not simply a reflection of the model’s inadequacy but an important depiction of the RMB’s operating mechanism. Combining time series segmentation with rolling window backtests further confirms the predictability boundary of the RMB exchange rate. Both validation methods, under different sample structures, exhibit highly consistent features. The model can extract low-frequency structural information but systematically underestimates spikes and rapid reversals in actual exchange rate movements. This recurring result across validation frameworks suggests that the limits of predictive ability are not due to sample segmentation or technical settings but stem from the intrinsic nature of exchange rate dynamics. Under a macro monthly information set, short-term fluctuations of the RMB exchange rate remain largely unpredictable, with a clear and stable upper limit on forecasting ability.

3. Ordered within intervals and unpredictable along paths. Building on this, we further provide a 12-month conditional forecast path. The results indicate that the RMB exchange rate exhibits typical chaotic evolution characteristics: forming an ordered interval under given constraints, but lacking predictability at specific points along the path. From a chaos theory perspective, exchange rates are not simply linear extrapolatable objects; their evolution involves both regularity and uncertainty. The systematic underestimation of spikes in out-of-sample predictions by the random forest model directly reflects this feature. Macro variables more easily constrain the central interval of the exchange rate but struggle to explain the magnitude and rhythm of short-term fluctuations. Comparing the conditional forecast with historical trends reveals that, under current macro conditions, the RMB is more likely to fluctuate mildly around a central appreciation rather than enter a one-sided trend. This characteristic aligns closely with the long-term operational logic of the RMB exchange rate.