This　paper　investigates　estimation　of　semiparametric　varying-coefficient　spatial　autoregressive　models　in　which　the　dependent　variable　is　missing　at　random.　An　inverse　propensity　score　weighted　sieve　two-stage　least　squares　(S-2SLS)　estimation　with　imputation　is　proposed.　The　proposed　estimators　are　shown　to　be　consistent,　no　matter　the　initial　value　is　taken　as　the　naive　S-2SLS　estimate　or　the　naive　nonlinear　least　squares　estimate,　and　the　asymptotic　distribution　of　the　latter　is　also　derived.　Simulation　studies　are　carried　out　to　investigate　the　performance　of　the　proposed　estimator.　The　method　is　finally　exemplified　with　one　real　data　set　on　Boston　housing　prices.