The Dictator’s Dilemma: To Punish or to Assist? Plan Failures and Interventions Under Stalin

43 Pages Posted: 19 Jul 2010

Date Written: September 20, 2007


The implementation of his decisions is one of dictator’s main concerns. The dictator knows that his orders may fail to be carried out because agents behave opportunistically, or because his orders contain mistakes, or both. The dictator is informed about which of his orders are implemented or fail, but not about the reasons for this. Imperfect information creates his dilemma: whether to punish an gent that acted poorly, or to assist her, or both. I suggest a simple game that models the dictator’s interventions. It predicts that under imperfect information the dictator’s choices depend on plans’ priority, i.e. the dictator’s utility associated with the expected return on their implementation. For high-priority orders the equilibrium strategy is to punish and assist simultaneously; where the expected return is sufficiently low a mixed strategy may be preferable. I verify the model against the case of the Stalinist command economy, using statistics of the decisions taken by the Communist Party's control commission (KPK) from 1934 to 1951, extracted from records in the formerly secret party archives. I find that KPK decisions coincided well with the logic of the game. This contributes to explaining random nature of punishment in dictatorships in general and underlines the importance of setting the problem of soft budget constraints in such systems in the context of the dictator’s coercive policy.

Keywords: dictatorship, principal-agent problem, soft budget constraints, USSR

JEL Classification: D73, P26, N44

Suggested Citation

Markevich, Andrei, The Dictator’s Dilemma: To Punish or to Assist? Plan Failures and Interventions Under Stalin (September 20, 2007). Available at SSRN: or

Andrei Markevich (Contact Author)

New Economic School ( email )

100 Novaya Street
Moscow, 143025
+79629851892 (Phone)
+74991293722 (Fax)


Here is the Coronavirus
related research on SSRN

Paper statistics

Abstract Views
PlumX Metrics