The idea is that the Princeton Bank Consortium blog delivers Bank/aggregate Sovereign numerical simulation results; portfolios, security models, market data, commentary, simulation code, and news interpretation relevant to bank balance sheet simulation. We cannot really do this through Pink Iguana because the Quant floating point curation is a much broader topic, in many ways.
Assume we have a 1000x floating point performance advantage over all banks worldwide just moving to -O2 compiled code and another factor of 20x going to optimized code with a contemporary optimizing compiler. We can run all USD assets/liabilities and break out each of the banks contributions with explanatories and match them against the data reported to the Fed in less time than the banks can compute their own data. The only difference is we do not have the Bank’s proprietary and mostly linear regression models ( not a massive deficit really). This has been true for at least 10 years and seems to have a ok chance or remaining true for another say 5 years. There is some room to maneuver and the Banks are not the biggest fans of rapid change. Moreover a Bank manager reporting to the Fed has almost no upside to doing it efficiently – any technology change is negative convexity to their budget and compensation. They are not P&L guys, they are Risk Professionals – they minimize risk.
We will work in machine efficiency space for clients since the performance is hard to track/understand. Yes, your cores are 100% utilized in top in your Java code doing software floating point emulation. We will have 5% efficiency Open source solutions and 50% and 98% efficiency optimized solutions. You pick – we will connect to Java, Erlang, Haskell, Clojure, OCaml, C++, A+/APL or your favorite language. It’s all good.
PBC will report on and update aggregated balance sheet simulation results quarterly with detailed interoperation of the explantories. The periodicity of reporting deadlines makes PBC a little like 538 or PEC for annual election cycles. The topic reporting will depend on the US Federal Reserve data and attempt to summarize cleanly like PredictWise, Kamakura, or Google Finance (as an aside: we could do this with Google it is the right size although it might be better in a solid comp. sci. grad school) . The blog will provide open source -O2 level simulation code but may use highly optimized code for publishing major results. We can do advertising for banks, Financial service firms, and microprocessors/cloud infrastructure.
538: Like how Silver runs Bayesian stat interpretation for national elections. Design PBC to run the same sort of data driven interpretation of national and international balance sheets. We like the Asset and Liability concentration in Big Banks since it makes the analysis and computation easier. It is a major Algorithmic playground. I’m not convinced we want PBC adding as much interpretive Bayesian sauce to the analysis as 538. Perhaps we stick closer the PEC’s data interpretation philosophy. Let’s see if there is a market for it one way or another. But we love Silver’s clean graphics, here, as a model for moving forward.
PBC will not have the resources to do anything like the news coverages 538 can do. But the video commentary is very good. Maybe we can do a poor man’s version of mstk3000 silhouettes reviewing the projected simulation results.
538 looks like they run from WordPress VIP – PBC could do that.
US Federal Reserve: Fed has the data to make USD NIMo possible. There is a massive amount of market and simulation data on a bank by bank basis. As a website the positives are the massive amount of data historical, current, and projected the site gives access to. The website is heavier than 538 or PEC but it churns much more content.
The Feds website menu structure is good for this content. Look at how they do Statistical releases, the FOMC Calendar, hooks to Twitter and Linked In. There is a Fed mobile app?
GLL: WordPress blog w. nice technical narrative structure. Very precisely targeted audience interacts through the well maintained comment section. They push some books and conferences but otherwise stay away from advertising more so than similar sites like DeLong’s. Picture and Prose format is good. the expertise for the blog topic is world class and is even top shelf among world-class Algorithm blogs. Brings color commentary to computational complexity – very novel. It’s like John Madden explains NP-completeness proofs with a grease pencil. These guys are way deeper in their field than Nate Silver is in his field, sort of like DeLong.
Princeton Election Consortium: Stronger in some ways than 538, but way less polished. This is Sam Wang’s side job that he runs with the assistance of a Princeton undergrad. Site has election cycle periodicity and pumps a good amount of data and scripts. The threading of the narrative with the data is great. Wang sells his Neuro books on the site, other than that little advertising is found. Lots of comments and Wang stays involved in the comments. Probably the best website of its class.
PredictWise: Very clean Tufte like data summary on the site – I want that for PBC. Nice menu structure as well with hooks to Twitter and Facebook. Like the clean data oriented first page with a menu link to the blog. PBC is selling Stochastic Simulation data first, standard product models second, computation third and then maybe interpretations.
Brad DeLong: Lipton/Regan, DeLong, Tao, and Gowers all kind of live in the same web site space of tight lucidity. DeLong is a little broader in topic selection. The narrative is top shelf. Reading lists, links, and lecture notes are real good. Politics colors the economic interpretation but that’s how the field is set up. Read the Chicago folks if it gets to you.
Yahoo Fantasy Basketball: Nice management of a relatively small amount of user data in the context of a season’s worth of stats for a fantasy league inside a professional sports league. Maybe run low end banks balance sheets and capital plan through a similar interface.
Chebfun: good model for pushing open source quantitative simulation code through a website. Looks like a Nick Trefethen operation.
Kamakura Corporation: Jarrow’s ( J in HJM) joint in Hawaii (in some ways he is much smarter than us). Very relevant content and services. Very not Open Software but these folks are all over CCAR, DFAST, BASEL, and risk computation. Very nice website and good layout.
Google Finance: Nice clean layout of elementary finance information. I think Marc Donner did this a few years back. Very clean but the narrative and the data are lite and very equity focussed.
nag blog: Library blog has nice services menu and standard layout. This was a Higham joint at some point, I believe.