Statistics · Regime Models
FIXED_THRESHOLD_REGRESSION
Fits separate OLS regressions above and below a threshold supplied by the user. Unlike THRESHOLD_REGRESSION, it does not search the sample for the best split.
Syntax
FIXED_THRESHOLD_REGRESSION(dataList, splitColumn, threshold)
Arguments
| Argument | Meaning |
|---|---|
dataList | A date-aligned DataSeriesList. Column 0 is the dependent variable; columns 1 onward are explanatory variables. |
splitColumn | Zero-based explanatory-variable column used to divide the sample. It must be 1 or greater. |
threshold | The finite numeric split selected by the analyst. Observations at or below it form one regime; observations above it form the other. |
Column convention: column 0 is always Y. The split variable is also included as an explanatory variable, and every regime regression uses all explanatory columns in the list.
One-line example
FIXED_THRESHOLD_REGRESSION(SAME_DATE_RANGE(LIST(1200*LOGDIFF(CPIAUCSL),UNRATE)),1,5)
This estimates annualized monthly CPI inflation as a function of unemployment, with separate regressions for UNRATE ≤ 5 and UNRATE > 5.
Returned analysis
- Fixed-threshold summary and regime observation counts.
- At-or-below and above-threshold slopes and t statistics.
- Pooled SSR, split SSR, SSR improvement and fixed-split F diagnostic.
- Actual-versus-fitted charts for both regimes.
- Regression tables for the two regimes and the pooled sample.
Extractable results
modelslideshowthreshold below_slopeabove_slope below_tstatabove_tstat pooled_ssrsplit_ssr ssr_improvementssr_improvement_pct f_statbelow_observations above_observationsbelow_model above_modelpooled_model
m=FIXED_THRESHOLD_REGRESSION(SAME_DATE_RANGE(LIST(1200*LOGDIFF(CPIAUCSL),UNRATE)),1,5) EXTRACT(m,above_slope)
Interpret carefully. Choose the threshold from theory, policy or a predetermined rule. A fixed split avoids threshold searching, but it does not create causal identification. Check that both regimes contain enough observations and examine residual dependence.
Related commands
THRESHOLD_REGRESSION searches for the SSR-minimizing split. MATCHED_COMPARISON forms balanced pairs on either side of a supplied group threshold.