Reference

MSCFit

Fitted multivariate synthetic-control model. theta is an n_controls x n_treated matrix mapping control outcomes to treated counterfactual outcomes.

MSCEstimate

Point estimate and fitted object for an MSC ATT analysis. Rows of counterfactual are post-treatment periods and columns are treated units.

MSCPanel

Balanced panel in matrix form. Rows 1:N0 are controls, rows N0+1:end are treated, columns 1:T0 are pre-treatment periods, and remaining columns are post-treatment periods.

fit_msc(Xpre, Ypre; kwargs...)

Fit the multivariate synthetic-control model from pre-treatment matrices.

Xpre is T0 x n_controls; Ypre is T0 x n_treated. The returned coefficient matrix maps controls to treated counterfactuals.

msc_estimate(Y, N0, T0; kwargs...)

Estimate ATT from an outcome panel matrix. Rows 1:N0 are controls, remaining rows are treated. Columns 1:T0 are pre-treatment periods.

predict_counterfactual(fit, Xpost)

Predict untreated counterfactual outcomes for treated units. Xpost is Tpost x n_controls; the result is Tpost x n_treated.

att(fit, Xpost, Ytreated_post)

Estimate ATT from a fitted MSC model and post-treatment outcomes.

objective(X, Y, theta, lambda)

Evaluate the MSC objective (1 / sqrt(size(X, 1))) * ||Y - X * theta||_* + lambda * ||theta||_1.

coef(fit)

Return the n_controls x n_treated MSC weight matrix.

panel_matrices(table, unit, time, outcome, treatment; treated_last=true)

Convert a balanced long panel to MSCPanel. Treatment must follow a common-adoption design: controls are never treated, treated units are untreated through T0 and treated in every post-treatment period.

simulate_msc(; kwargs...)

Generate a low-dimensional synthetic-control simulation with many treated units. Returns a named tuple with Y, N0, T0, true theta, untreated counterfactuals, and metadata.

fitted_pre(fit, Xpre)

Return fitted treated-unit outcomes in the pre-treatment period.

residuals(fit, Xpre, Ypre)

Return pre-treatment residuals Ypre - fitted_pre(fit, Xpre).

prefit_rmse(fit, Xpre, Ypre)

Root mean squared pre-treatment fitting error.

effect_matrix(est)

Return post-treatment effects by period and treated unit.

effect_curve(est)

Average treatment effect by post-treatment period.

unit_effects(est)

Average treatment effect by treated unit.

weights_table(fit; units=nothing, treated_units=nothing)

Return the estimated weight matrix as a DataFrame with one row per nonzero control-treated pair.

cv_table(fit)

Return cross-validation errors by lambda. If the model was fit without CV, the error column is missing.

placebo(est; T0_placebo=nothing, kwargs...)

Move the treatment date into the pre-treatment period and estimate a placebo ATT.

placebo(Y, N0, T0; T0_placebo=floor(Int, T0/2), kwargs...)

Estimate a time-placebo effect using only the original pre-treatment periods.

placebo_distribution(Y, N0, T0; replications=100, ntreated=nothing, kwargs...)

Randomly assign control units to placebo treatment groups and estimate MSC effects using the original treatment date. This is most useful when N0 is much larger than the number of treated units.

placebo_se(Y, N0, T0; kwargs...)

Standard deviation of placebo_distribution.