The post-pandemic surge in inflation raised some questions as to which measure of core inflation was the best indicator of future total inflation, the Bank of Canada’s official target. At that time we found that CPI excluding food, energy and the effect of indirect taxes (CPIXFET) outperformed other measures while having the added benefit of being more intuitive, and easier to communicate to markets and households.

Now, with inflation largely back within the Bank of Canada range, we’ve repeated this exercise, comparing CPI-trim, CPI-median, and CPIXFET. Once again, CPIXFET proves to be the most accurate predictor of headline inflation.

Our approach involves estimating augmented Phillips Curves for each of the core measures, including one for the average of CPI-trim and CPI-median. Each curve links year-over-year inflation to economic slack and other fundamentals. We then estimate bridge equations that connect each core measures to total inflation. This allows us to assess which core forecast best predicts headline inflation.

To do so, we run dynamic simulations that jointly forecast core and total inflation over history. This is a more stringent test of the forecasts’ performance than an in-sample approach, which reset forecast errors at each period based on observed inflation.

Table 1 presents the estimation results of the augmented Phillips Curves, while table 2 reports the corresponding results for the bridge equations. Both sets of equations are estimated using the Generalized Method of Moments within a semi-structural general equilibrium framework that links inflation dynamics to economic fundamentals and incorporates the monetary policy response through a Taylor rule.

Table 1: Phillips Curves Estimation Results
Table 2: Total CPI Bridge Equations Estimation Results

Because total inflation forecasts are derived from those of core inflation, we assess predictive performance by comparing the root mean squared errors of total inflation forecasts. Table 3 reports forecast accuracy scores across different simulation periods. Over the 2018–2025 period, the CPIXFET is the clear winner based on RMSE results. Chart 1 illustrates the dynamic simulation forecasts for total inflation over this period, providing a visual sense of how each measures track headline inflation. 

Table 3: Root Mean Squared Errors of Bridge Equations Over Different Simulations Periods
Chart 3: Dynamic Simulations of Headline Inflation Versus Actual

During the inflation rundown starting in 2022Q3, all core measures perform similarly, though CPIXFET delivers the best overall forecast of total inflation when considering the whole sample. Notably, only CPI-Trim captures the peak of the inflationary episode, albeit at the cost of generally overestimating inflation leading up to and following the peak.

While CPIXFET maintains the lead in the 2021–2025 simulation (table 3 again), its advantage over CPI-trim and CPI-median narrows. When the model is simulated over the most recent eight quarters, CPI-trim achieves the most accurate forecast (lowest RMSE), followed by CPI-median, with the average of the two and CPIXFET performing similarly thereafter. Importantly, the core-inflation-based equations outperform a random walk across all simulation periods.

Although shorter-sample simulations provide interesting insights, the 2018–2025 sample fully captures the most recent inflationary cycle, including multiple shocks such as supply constraints and geopolitical tensions. The strong performance of CPIXFET over this broader horizon simulation reinforces its value as a reliable guide for predicting total inflation and informing monetary policy decisions, especially during times of heightened uncertainty.