||This paper investigates empirically the effect of real assets as collateral on the economy. As pointed out by Kiyotaki and Moore (1997), when loans are collateralized and firms are credit constrained, the amount borrowed is determined by the value of the collateral; a decrease in the price of productive assets will therefore have a negative impact on firm investments. These effects are cumulative, which leads to credit cycles. There are potentially two major difficulties in empirically testing the relationship between collateral and firm investments. First, the value of collateral often is not observable due to the lack of active secondary markets for collateralizable assets, such as plants and machineries. Second, collateral is endogenous. For example, when firms invest they need to purchase machines and build plants, which expands their collateralizable assets. This paper deals with these two difficulties by using the land-price collapse in Japan in the early 1990s as a natural experiment. In Japan, the main form of collateral for corporate borrowing is land, the value of which is observable. In addition, between 1990 and 1993, there was a near 50% drop in land prices, which was unambiguously exogenous to any one individual firm. As firms suffer losses proportionate to their land holding, the amount of land held prior to the shock can serve as an exogenous instrument to measure collateral. I find that collateral affected Japanese firm investments in two important ways. The first is a collateral-damage effect: losses in collateral value reduced investments. The second is an indirect internal-liquidity effect: with reduced borrowing capacity, the firms had to rely more upon internally generated cash to finance their investments. In addition to the investment analysis, we use a unique dataset of matched firm-bank lending, which permits us to further examine the mechanism through which collateral affects firm investments. In particular, I investigate whether collateral losses also leads to reduced borrowing capacities. The matched sample allows me to control for unobservable heterogeneity in the loan supply. The results show, again, a significant collateral-damage effect: banks tended to lend less to those who suffered greater collateral losses.