I estimate a network-based model of OTC markets developed in Gofman (2011) by using federal funds market data to study the trade-off between efficiency and stability of different financial architectures. The estimated financial architecture with a small number of large interconnected banks is 11 times more efficient than a regulated financial architecture of the same size and density but without these institutions. The estimated architecture is more efficient because it requires fewer intermediaries to allocate the same liquidity shocks. In addition, large interconnected banks have more bargaining power and improve efficiency even more. However, a failure of the most interconnected bank that triggers a cascade of defaults shows that during extreme shocks the estimated architecture is more fragile than the counterfactual one. The number of surviving banks is 30% higher in the regulated architecture when the risk of contagion is high. Overall, the proposed framework allows us not only to estimate the structure of trading relationships in an OTC market based on a network of observed trades, but also it allows us to quantify the efficiency and stability of the current financial architecture and an alternative architecture that arises because of the regulation of too interconnected to exist banks.