Thanks for your reply!
However, I don’t think the send-to-self attack could cause serious problems.
First, the send-to-self attack is costly since the attacker needs to lock lots of capital in his “fake” project. Meanwhile, his return will be marginal. Suppose the attacker makes a project that only he invests and no other investors invest. It is true that by rising his investment in this project, he can increase the estimation of correlation between him and other investors. However, the subsidy that he can extract from his “fake” project is still 0. For other projects, increase in subsidies caused by the attack should be very limited and such increase can’t be enjoyed solely by the attacker.
Second, in calculating the discoordination coefficient, we can simply exclude any project with only one investor. For example, we can consider those projects to be failed in fundraising and not qualified for QF subsidies. In fiat denominated crowdfunding, there is often a threshold mechanism: Any project that is unable to win enough supporters or capital commitment within a certain time frame will be considered as failed, and any capital commitment will be returned thereafter. I think QF can incorporate similar mechanism.
Third, we can use data from past rounds of QF to estimate the discoordination coefficient and make its estimation less impacted by a single data point.
For the “higher-order measures of correlation”, I have done some preliminary research. Basically, besides correlation among different investors, we should also consider correlation among different projects. Just like coordinated investors tending to make similar investment decisions, related projects tend to attract similar groups of investors. When using investment data to estimate correlation among investors, we should be aware of the interdependence among projects.
I believe Singular Value Decomposition (SVD) is the right tool to study this problem. However, I think the solution may be too complicated to communicate to the community of investors. Besides, the solution may be susceptible to over-fitting.