Sandpile economics, recast as a same-game parlay
"Sandpile Economics: Theory, Identification, and Evidence" argues that financial and economic networks quietly build up geometric curvature — a Forman–Ricci / Ollivier–Ricci style measure on the graph of who's connected to whom — and that the parts of the network with low or negative curvature are structurally poised for disproportionate cascades, even when they don't look unusually connected by raw degree. A sandpile looks perfectly stable right up until one more grain lands in the wrong spot; the paper's claim is that curvature, not connection count, tells you where that grain is. arXiv:2604.13890 →
Everything below is our own analogy, not the paper's method. We recast "nodes" as same-game parlay legs and "edges" as correlation between legs (same team, same game, correlated props), then compute a simplified weighted Forman–Ricci-style curvature score live in your browser — sportsbooks are frequently accused of pricing correlated parlay legs closer to independent than they really are. The exact formula, odds, and thresholds here are illustrative, not from the paper, and not betting advice.
Drag a leg to rearrange. Click a leg (without dragging) to shock it — simulate it missing — and watch the cascade. Node size = naive connectivity (degree). Node color = curvature (red = fragile, green = robust).
Each correlation edge carries a weight w between 0 and 1 (how tightly two legs move together). For an edge e = (u, v) we use a weighted Forman–Ricci curvature (node weights fixed at 1): F(e) = 2 − √w · (Σ 1/√wi over u's other edges + Σ 1/√wj over v's other edges). Edges sitting inside dense, mutually-reinforcing clusters — e.g. LeBron's scoring prop next to the game total next to the Lakers spread, all reinforcing each other — score high curvature. Thin edges with little shared structure, especially bridges between otherwise unrelated legs, score low or negative curvature: the fragile links the paper flags. A leg's own curvature is the average of its incident edges. This is a simplified, triangle-free variant of Forman–Ricci curvature — illustrative of the paper's idea, not a reproduction of its estimator.