This paper develops a Markov-switching vector autoregressive (MS-VAR) model with time-varying transition probabilities to model downside risks to economic growth in the Growth-at-Risk (GaR) framework. While GaR models traditionally rely on quantile regressions or factor-based approaches to capture the influence of financial conditions on the lower tail of growth, I adopt a regime-switching specification in which the probability of entering a high-risk regime is itself driven by a large set of macro-financial predictors. Sparsity is induced through an ℓ1-penalized maximum likelihood estimator, allowing the data to identify the most relevant indicators for regime changes without pre-selecting variables or aggregating them into factors. Using U.S. data, the model uncovers that credit spreads, measures of leverage, and uncertainty indices are systematically selected as key transition drivers, and that regimes associated with elevated downside risks coincide with periods of financial stress and tight credit conditions. By integrating variable selection into the transition mechanism, the model delivers a structural and data-driven map of the conditions that generate shifts in growth risk, providing a complementary perspective to existing GaR methodologies.