the possibility of a bank run, such an intermediary can in effect exploit the law of large numbers to allow each investor to withdraw money at will, while still having a predictable aggregate withdrawal in period 2. To see the advantage of this pooling, suppose that the intermediary were to allow each contributor of capital to withdraw I+r in period 2, and suppose that the intermediary knows that a fraction p of the population will turn out to be period-2 consumers. Then all the intermediary needs to do is put a fraction p of the funds it receives into the short-term asset, l-p into the long-term asset then each investor will expect to receive 1+r if he consumes in period 2, (1+h) if he consumes in 3 dominating the range of possibilities available without the intermediary. In general, of course investors will choose some other point, such as B, on the budget constraint passing through that point, so that they will do even better So far so good. But such an intermediary is, as Diamond and Dybvig pointed out, potentially subject to a bank run. In our case this possibility arises because the liquidation value v is less than the promised payout to early withdrawers The point is straightforward. Suppose that for some reason- it does not matter what that reason is investors who would ordinarily not have withdrawn their funds become convinced that many other such investors will attempt to withdraw their funds. Should investors who plan to consume in period 3 nonetheless withdraw funds in period 2, the intermediary will not have enough of the safe asset, and will therefore have to liquidate projects in mid-stream; and since the liquidation value is less than the promised payout, not all investors will in fact be able to withdraw their funds. And therefore the rumor of such a run will lead to a rush to withdraw funds by investors anxious not to be last in line (Of course, in principle the possibility of a run should be taken into account in the initial investments and offers by the intermediary one can justify the approach here by supposing that such a run is perceived as a very unlikely event) As in the previous model, this gives us a story about a crisis that can be set off merely by self- fulfilling expectations. However, in the moral hazard model the pre-crisis state of affairs is fundamentally unsustainable; in effect, the asset market is in a"metastable"state, like a sandpile with a"supercritical"slope, and any small shock causes an avalanche -a slump in asset values toward their appropriate level. In the bank-run model, the state pre-crisis was reasonable, and could have been sustained, but was undermined by an unnecessary panic -which produces real costs due to the premature interruption of productive activities Both views can be given some support from anecdotal evidence, as argued below. But let us turn next to the implications of the two views for foreign direct investment 4. The role of fdi As Kindleberger(1969)pointed out long ago, foreign direct investment is essentially about transfer of control rather than movement of capital pe Indeed, a quick look even at balance-of-payments measures of FDI for emerging market economies reveals that there is very little relationship between overall capital flows and FDI. Tables I and 2 show overall capital inflows and inward FDI for Argentina and Mexico from 1990-96; even though such balance of payments numbers tend to confuse internal capital transfers within firms(which behave like portfolio capital) with true changes in control, there is still a striking lack of correlation -or perhaps even an inverse correlation between overall capital inflow and FDIthe possibility of a bank run, such an intermediary can in effect exploit the law of large numbers to allow each investor to withdraw money at will, while still having a predictable aggregate withdrawal in period 2. To see the advantage of this pooling, suppose that the intermediary were to allow each contributor of capital to withdraw 1+r in period 2, and suppose that the intermediary knows that a fraction p of the population will turn out to be period-2 consumers. Then all the intermediary needs to do is put a fraction p of the funds it receives into the short-term asset, 1-p into the long-term asset; then each investor will expect to receive 1+r if he consumes in period 2, (1+h)2 if he consumes in 3 - dominating the range of possibilities available without the intermediary. In general, of course, investors will choose some other point, such as B, on the budget constraint passing through that point, so that they will do even better. So far so good. But such an intermediary is, as Diamond and Dybvig pointed out, potentially subject to a bank run. In our case this possibility arises because the liquidation value v is less than the promised payout to early withdrawers. The point is straightforward. Suppose that for some reason - it does not matter what that reason is - investors who would ordinarily not have withdrawn their funds become convinced that many other such investors will attempt to withdraw their funds. Should investors who plan to consume in period 3 nonetheless withdraw funds in period 2, the intermediary will not have enough of the safe asset, and will therefore have to liquidate projects in mid-stream; and since the liquidation value is less than the promised payout, not all investors will in fact be able to withdraw their funds. And therefore the rumor of such a run will lead to a rush to withdraw funds by investors anxious not to be last in line. (Of course, in principle the possibility of a run should be taken into account in the initial investments and offers by the intermediary; one can justify the approach here by supposing that such a run is perceived as a very unlikely event). As in the previous model, this gives us a story about a crisis that can be set off merely by self￾fulfilling expectations. However, in the moral hazard model the pre-crisis state of affairs is fundamentally unsustainable; in effect, the asset market is in a "metastable" state, like a sandpile with a "supercritical" slope, and any small shock causes an avalanche - a slump in asset values toward their appropriate level. In the bank-run model, the state pre-crisis was reasonable, and could have been sustained, but was undermined by an unnecessary panic - which produces real costs due to the premature interruption of productive activities. Both views can be given some support from anecdotal evidence, as argued below. But let us turn next to the implications of the two views for foreign direct investment. 4. The role of FDI As Kindleberger (1969) pointed out long ago, foreign direct investment is essentially about transfer of control rather than movement of capital per se. Indeed, a quick look even at balance-of-payments measures of FDI for emerging market economies reveals that there is very little relationship between overall capital flows and FDI. Tables 1 and 2 show overall capital inflows and inward FDI for Argentina and Mexico from 1990-96; even though such balance of payments numbers tend to confuse internal capital transfers within firms (which behave like portfolio capital) with true changes in control, there is still a striking lack of correlation - or perhaps even an inverse correlation - between overall capital inflow and FDI