Dynamic Risk Preferences, Loss Chasing, and Reference-Dependent Stopping in Repeated Contests: Field Evidence from Online Poker

We examine how dynamic within-session wealth fluctuations alter risk-taking and session termination decisions in repeated contests. Analyzing 4,070,556 hand-level decisions across 119,584 completed sessions from the University of Alberta Computer Poker Research Group archive (1995-1997), we exploit continuous bankroll tracking within an environment where objective probabilities are governed by invariant combinatorial rules. Controlling for player fixed effects, table stakes, position, and time-on-task, we identify two distinct behavioral regimes. In the gain domain, competitors exhibit a house money effect: accumulated profits increase voluntary pot entry (VPIP, +0.59 percentage points, p < 0.001), preflop raising (+0.27 percentage points, p = 0.003), and absolute wager sizes (+0.57 chips per 100 chips gained, p < 0.001). In the loss domain, increasing cumulative deficits trigger systematic loss chasing: holding current stack size constant, each 100 chips of deficit increases absolute bet sizes by +0.99 chips (p < 0.001) and escalates the bet-to-stack ratio (t = 15.52, p < 0.001). At the session level, termination choices exhibit sharp reference dependence. Players display an acute "never quit on a win" asymmetry: the per-hand exit hazard is 0.36% immediately following a winning hand, compared to 2.48% after a loss and 5.02% after uncontested folds. Furthermore, 1,340 sessions terminate at exact net break-even (W = 0), with active round-trip clawbacks accounting for 54.9% of these exits. While capital constraints truncate losing sessions on average, surviving losing sessions display pronounced right-tail persistence, consistent with selective sunk-cost chasing.