This study examines the bank-level determinants of profitability in India using a balanced panel of 20 scheduled commercial banks (10 public, 10 private) over 2005–2025 (420 bank-years) drawn from Reserve Bank of India statistics. Because the number of banks is small relative to the number of years, we propose a transparent model-selection protocol and adopt a dynamic two-way fixed-effects estimator as the primary model, with bank-clustered, Driscoll–Kraay and wild-cluster-bootstrap inference, and two-step system GMM with collapsed instruments as a complementary check. Net non-performing assets (NPAs) are the dominant drag on return on assets (short-run coefficient −0.177; long-run −0.236), and the effect is convex, weakening at high NPA levels. Net interest margins are positively related to profitability (0.259), whereas the credit–deposit ratio shows no robust association. Profits are moderately persistent (φ between 0.25 and 0.34). The capital adequacy effect is positive but sensitive to specification and inference. The pass-through of margins to profits is significantly stronger after 2016 (joint F-test p = .002). The NPA penalty is larger for private than for public banks, but the ownership difference is statistically significant only without year effects (p = .041 versus .237 with them). Panel Granger tests indicate feedback between NPAs and profitability, so the estimates are conditional associations rather than causal effects. The findings inform asset-quality supervision and capital policy in bank-dominated emerging economies.