I model people in a coordination game who use a communication network to tell each other their willingness to participate. The minimal sufficient networks for coordination can be interpreted as placing people into a hierarchy of social roles or "stages": "initial adopters", then "followers", and so on down to "late adopters". A communication network helps coordination in exactly two ways: by informing each stage about earlier stages, and by creating common knowledge within each stage. We then consider two examples: first we show that "low dimensional" networks can be better for coordination even though they have far fewer links than "high dimensional" networks; second we show that wide dispersion of "insurgents", people predisposed toward participation, can be good for coordination but too much dispersion can be bad.

Using a comprehensive list of terrorist attacks over three decades, we find that aggregate investor risk aversion inversely relates to terrorist activity in the United States. A one standard deviation increase in the number of attacks each month leads to a 75.09 million drop in aggregate flows to equity funds and a 56.81 million increase to government bond funds. Tests on alternative channels further suggest that the shift in aggregate risk aversion is driven mainly by an emotional shock rather than changes in wealth or the outside environment. We also investigate possible alternate explanations for reduced flows to risky assets. Our evidence is consistent with a fear-induced increase in aggregate risk aversion.