How Google's algorithm could fix the financial system
Computation could make banks’ systemic risks visible to all
About 15 years ago, Google's PageRank algorithm transformed the web by making it much easier to search. An algorithm, cleverly applied, reconfigured a sea of information and transformed the world. That same algorithm, transported and translated into the world of banking, could do the same for our perpetually unstable financial system.
That's the message of a little known body of work developed by some physicists and economists over the past few years. The best way to eliminate the problem of "too big to fail," they suggest, isn't through complex regulations that banks will inevitably find ways to work around. Rather, it's to design the system to have automatic radical transparency, so that financial institutions have natural incentives to act in ways that improve overall system stability, even as they go about seeking their own profits. In helping themselves, they will also help the system.
A variant of Google's algorithm would provide the core tool to make it happen. Here's how:
Two years ago, some physicists introduced a new measure of systemic risk they called DebtRank, which took its inspiration from the famous PageRank algorithm invented by Google founders Sergey Brin and Larry Page. The essential insight of the PageRank algorithm is that in any network of things that make references to each other -- web pages through hypertext links, or scientific papers through citations, for example -- each element effectively votes for the importance of other elements by linking to them. Hence, the most important web pages (or papers) should be those drawing links from many other pages (or papers), especially from other really important ones.
To calculate the PageRank of a web page you have to look at all the web pages that link to it. The page gets a higher PageRank in so far as many other pages link to it, especially if those other pages are important, i.e. are pages that have many other web pages linking to them, especially other important web pages. The definition is obviously circular -- you have to know which pages are important in order to be able to calculate which pages are important. That may seem useless, but this problem is easy to sort out mathematically (it's just some linear algebra) and that's what Google algorithm does using lots of computing power.
Now, what about finance? The analogy for DebtRank is quite direct -- those institutions that present the greatest risks to the financial system are those that, if they fail, would cause the widest spread of economic distress. Naturally, you would tend to have a high DebtRank if you are linked by loans and other financial ties to other firms with high DebtRank -- the same circularity again. I won't go into more detail except to say that DebtRank can be calculated in a straightforward way using knowledge of the financial links between institutions (information held, partially, by Central Banks). It's a method for exposing the true systemic risks in a way that is currently not done.
Now, how can that be used to make the system safer? Very recently, two physicists in Austria showed how DebtRank could be used to design a very clever tax on financial transactions. The tax would be specifically linked to how much systemic risk a transaction -- say a loan from one bank to another -- creates. It's easy to roll your eyes when you hear about transaction taxes, thinking first that it will probably never happen, and second that it might not do much good even if it did. But this new idea is VERY different from anything ever proposed before. It uses computation to do something that was never possible before.
This would be, in fact, a tax that financial institutions WOULD NOT HAVE TO PAY. Institutions that work hard to borrow and lend in a way that doesn't increase risk to the overall financial system (by piling up debt on particular institutions, for example) would end up paying no tax. The idea is to bring systemic risk into the pricing system so institutions have an incentive to avoid it. In so doing, you provide a mechanism for the entire financial network to reconfigure itself to have lower systemic risk. The paper proposes a concrete method to do this, although it would in practice require giving central banks more information on financial transactions of many kinds. I’ve given some further detail on how it might work here.
Does anything like this have a chance? I happened to speak to some people in the IMF and the OECD last week at a conference and they said, in the short term, sadly, no. It will be an uphill battle first because ideas like this seem strange to economists and don't fall into the framework they use to think about financial systems, and second because the big banks don't really want any mechanism to reveal just how risky they are to the system. The biggest banks make their profits precisely by taking on such risks, pocketing the gains, and sticking taxpayers with the costs when things go bad. They don't want anything like radical, algorithmic based systemic risk transparency.
But the rest of us should. Currently, we're the ones paying for those risks.
Follow The Physics of Finance on Twitter: @Mark_Buchanan