Richard Fairbank, Capital One Financial

Richard D. Fairbank is the CEO of Capital One Financial Corporation (COF). COF reported strong third quarter earnings yesterday. The number I was waitbank-richard-fairbank-900cs052313-e1442371136993-1024x670ing for was the updated Net Interest Margin which came in at 679 basis points in the presentation for 3Q2016 Earnings. The NIM has been relatively flat for the past year at COF with moves of around 6 bps max quarter over quarter. The reason this NIM number is so interesting is that it is at least two times larger than the NIM at JPM (oddly, does not actually quote a NIM number other than to say in a footnote it is flat QoQ), Citi (286 bps), or BofA (223 bps). I recently wrote a paper titled What’s in their Wallet discussing this observation. Fairbank appears to have quantitatively optimized parts of the COF capital plan implementation process while his competition runs a manual process. If banking books are like rockets, COF is using computer guidance while the competitors are mostly steering with a joystick.

There are two reasons this is interesting. First, Financial engineering is different than other numerical/computational engineering fields like fluid mechanics, light diffraction, or supersonic flow. In Finance there is no underlying physical reality  for the models to calibrate, it is all business convention. Twenty years ago nearly all the quant models prohibited negative interest rates in any expectation calculation. Now  negative yields on global government bonds is a reality and the quant texts are being rewritten.  When a bank like COF establishes a quantitative methodology like NIM Optimization in the market by adding ~$10B in annual revenue they are doing their part to establish a new Financial engineering reality.  Second, the time has come for banks to model each of the securities and contracts on the banking book just the same way as they model trades on the trading book. Sure a large bank might have a few billion positions in its accrual portfolio but with current technology that is a modest computational challenge. Moreover, with new FinTech these billions of securities are going to become computational devices like phones. Fairbank is simply ahead of the game in assembling the infrastructure to automate banking in this new FinTech world.

Net Interest Margin

The Net Interest Margin (or NIM) is one of the primary measures of a Bank Holding cropped-fredgraph4.pngCompany’s (BHC) capacity to generate revenue. NIM figures reported by individual banks can be found at  BankRegData.com and in aggregate at the St Louis Fed as displayed here in the FRED graph. The FRED graph shows the average NIM of the U.S. Bank Holding Companies. Notice that average NIM is close to a 30-year low at 300 basis points (or 3%). That means the difference in the lending rates (assets) and the deposit rates (liabilities) are  close to a minimum seen over the past 30 years.  The aggregate US commercial bank assets is 15.96T on 21 Oct. 2016 and the total reported liabilities  is 14.21T on 12 Oct. 2016. Nearly half of assets ~7.3T reside in a handful of the largest BHCs: JPM, Wells. BofA. Citi, US Bancorp, and PNC. For a bank with the assets the size of JPM a 1 basis point (bp) increase in the NIM generates an additional ~$200MM  annual revenue.

Numerous factors drive the NIM level, perhaps the most important one is that at the end of the business day the value of the BHC’s assets and liabilities+equity  must be equal according to GAAP, IASB, and FASAB. This is also referred to informally as the “assets must be funded.” The bank may use transfer pricing to allocate capital within the firm so that funding assets may be broken down into several smaller processes. For example, a global bank might use transfer pricing for capital allocation between Asia, Western Europe, South America, and North America.  Finally, the bank generally has a capital allocation plan for new funds and excess cash flow from the banking book (or accrual portfolio). The capital plan may also allocate funds for hedging the interest rate, credit, and FX risk in the banking book.  The timing and selection of acquiring assets and liabilities, maintaining new capital reserve levels, the efficiency of the capital allocation plan implementation in light of market movements, as well as the market performance of the roll-over accrual portfolio inventory  are important factors determining the NIM level.

 

Beat the Dealer

If you had to pick one person who made it possible to analyze and stochastically im3-1optimize all the Assets and Liabilities of U.S. Bank Holding Companies, a leading candidate in any such discussion is Edward O. Thorp. In 1966 he published the book Beat the Dealer explaining how to “count cards” to beat the house (the dealer) in the casino card game Blackjack. Thorp famously used an IBM 704 mainframe computer running Fortran programs to solve Blackjack. What is less commonly known is that Thorp’s applied mathematics research for his hedge fund Princeton/Newport Partners very likely had a early version of the Black Scholes equation to use for successful proprietary trading for several decades starting well before the publication of the Nobel Prize winning 1973 Black Scholes paper. Thorp is one of the earliest and most successful practitioners of computational finance in a global market over a sustained time period.

There are many other seminal contributors in the discussion: Chebyshev, Dantzig, Metropolis, Kahan, Moore, Bellman, Stallman, Raineri, Knuth, and Cray come to mind. In this blog we will explore the connections and trace the evolution of their foundational ideas into current research in this corner of computational finance and applied mathematics.

About

img_0018I am Jonathan Sandberg and I run the Princeton Bank Consortium. I came to Wall Street from Princeton University in the 90s. I have worked on the trading floor in Morgan Stanley DPG, Lehman Brothers Credit, Citi FX, and Goldman PB High Frequency Equities. I am a floating point inner-loop hacker  for exotic derivatives, large portfolios, and low-latency trade execution (see Black Scholes in 36 nanoseconds, Totally Serial, or the 2012 Due Diligence series on Maxeler/JPM Credit Derivative Supercomputer).

The Princeton Bank Consortium (PBC) will be a data publication outlet for a new type of Big Data stochastic programming problem focussed on the optimization of U.S. Bank Net Interest Margin. Stochastic Programming is a well researched industrial and academic topic see Shapiro, Shapiro et.al., or Nemirovski. There is even solid practical experience in Stochastic Programming applied directly to Bank Asset and Liability Management see Birge, QRM, Oracle Financial, and IBM Algo. What has changed within the past couple years is that the Banking industry efforts in automation, centralization, and regulatory reporting have made Net Interest Margin Optimization feasible on a massive scale. New Technology including: inexpensive vectorized x86 microprocessors, fast LP solvers, and solid optimizing compilers make it so PBC can run Net Interest Margin Optimization for the aggregated accrual portfolio of JPM, Citi, Wells Fargo, BofA, and the remaining top 50 Bank Holding Companies (BHCs) tracked by the U.S. Federal Reserve on an iMac with open source software (of course PBC does not know the exact accrual portfolio contents of each of these BHCs, but there is aggregated Fed data to work with). This is a very exciting, and lucrative, new bank automation quant-problem that the Princeton Bank Consortium will assist in promoting, finding, and optimizing industrial solutions.

This PBC blog will be used to provide the narrative for the development of Net Interest Margin Optimization solutions. The blog will present the people and the research that makes solutions possible and trace the evolution of the algorithms, programs, and systems that make progress in NIM Optimization possible.