Evaluate inflation forecasts

Runs standard forecast evaluation tests. The bias test (Mincer-Zarnowitz) checks whether forecasts are unbiased. The efficiency test (Nordhaus) checks whether forecast errors are autocorrelated. The Diebold-Mariano test compares predictive accuracy of two competing forecasts.

Usage

ik_forecast_eval(
  actual,
  forecast,
  test = c("bias", "efficiency", "dm"),
  forecast2 = NULL,
  horizon = 1L,
  alternative = c("two.sided", "less", "greater")
)

Arguments

actual

Numeric vector. Realised inflation values.

forecast

Numeric vector. Forecast values, same length as actual.

test

Character. One of "bias", "efficiency", or "dm".

forecast2

Numeric vector or NULL. Second forecast series (required for "dm" test), same length as actual.

horizon

Integer. Forecast horizon (used for Newey-West bandwidth in the DM test). Default 1.

alternative

Character. Alternative hypothesis for the DM test: "two.sided", "less", or "greater".

Value

An S3 object of class "ik_forecast_eval" with elements:

test

Character. The test performed.

statistic

Numeric. The test statistic.

p_value

Numeric. The p-value.

coefficients

Named numeric vector (for bias and efficiency tests).

conclusion

Character. A plain-language summary of the result.

References

Diebold, F. X. and Mariano, R. S. (1995). "Comparing Predictive Accuracy." Journal of Business and Economic Statistics, 13(3), 253-263.

Examples

set.seed(42)
data <- ik_sample_data("headline")
actual <- data$inflation[5:80]
forecast1 <- data$inflation[4:79] + rnorm(76, 0, 0.2)
forecast2 <- data$inflation[4:79] + rnorm(76, 0, 0.4)

# Mincer-Zarnowitz bias test
bias <- ik_forecast_eval(actual, forecast1, test = "bias")
print(bias)
#> 
#> ── Forecast Evaluation ─────────────────────────────────────────────────────────
#> • Test: Mincer-Zarnowitz Bias Test
#> • Test statistic: 19.5379
#> • p-value: 0
#> • Reject H0: forecasts appear biased (alpha != 0 or beta != 1).
#> • Coefficients: alpha = 0.7846, beta = 0.4673

# Diebold-Mariano test
dm <- ik_forecast_eval(actual, forecast1, test = "dm", forecast2 = forecast2)
print(dm)
#> 
#> ── Forecast Evaluation ─────────────────────────────────────────────────────────
#> • Test: Diebold-Mariano Test
#> • Test statistic: -3.3516
#> • p-value: 8e-04
#> • Forecast 1 is significantly more accurate than Forecast 2.
#> • Coefficients: mean_loss_diff = -0.126