Reinventing Bookkeeping and Accounting (In Search of Certainty)

First draft. Feedback, criticisms, ideas and links to related works would be greatly appreciated.

Double-entry bookkeeping was deployed in its modern form in the 1300s. While minor innovations have occurred since then, the fundamental atomic unit of tracking and managing value–our accounting system–is still based on this 700-year-old invention. With today’s computers, networks, and cryptography, we now have the opportunity to create a system of accounting that brings us into the 21st century–a system that looks beyond numbers in ledgers and utilizes machine learning, multiparty computation, and algorithmic representation to redefine “value.” So what’s holding us back?

Accounting underlies finance, business, and enables the levying of taxes for raising armies, building cities, and managing resources at scale. In fact, it is the way that the world keeps track of almost everything of value.

Accounting predates money, and was originally used by ancient communities to track and manage their limited resources. There are accounting records from Mesopotamia dating back more than 7,000 years, listing the exchange of goods. Over time, accounting became the language and information infrastructure for trade. Accounting and auditing enabled the creation of vast empires, such as those built by the Egyptians and the Romans.

As accounting scaled, it made sense to go from counting sheep, bushels of grain, and cords of wood, to calculating and managing resources using their exchange value in terms of an abstract unit: money. In addition to exchange, money allowed for recording and managing obligations. So where earlier bookkeeping just kept records of promises and exchanges between individuals (Alice lent Bob a goat on this date), money opened up a new realm of accounting by dramatically simplifying the management of accounts and allowing markets, companies, and governments to scale. However, through the centuries, this once powerful simplification has a resulted in a surprising downside–a downside made worse in today’s digitally connected world.

Defining Value

While companies today use enterprise resource planning (ERP) systems to keep track of widgets, contractual obligations, and employees, the accounting system–and the laws that support it–require us to convert just about everything into monetary value, and enter it into a ledger system based on the 700-year-old double-entry bookkeeping method. This is the very same system used by the Florentine merchants of the 13th century and described by Luca Pacioli, the “father of accounting,” in his book Summa de Arithmetica, Geometria, Proportioni et Proportionalità, published 1494.

When you take, for instance, a contract that pays out $1 million if it rains tomorrow, and put it into your accounts, you will be required to guess the chance of rain–maybe 50 percent–and value that asset at something like $500,000. The contract will actually never pay out $500,000; in the end, it will either be worth zero (no rain) or $1 million (rain). But if you were forced to trade it today, you’d probably sell it for something close to $500,000; so for tax and management purposes, you “value” the contract at $500,000. On the other hand, if you are unable to sell it because there are no buyers, it might actually be valued at zero today by regulators interested in liquidity, but then suddenly valued at $1 million tomorrow if it rains.

Basically, a company’s accounts are an aggregate of cells in various ledgers with numbers that represent a numerical value denominated in some currency–yen, dollars, euros, etc.–and those numbers are added up and organized into both a balance sheet and an income statement that show the health of the company to management and investors. They are also used to calculate profits and the amount of tax owed to governments. This balance sheet is a list of assets and liabilities. If you looked in the assets column, you’d have a number of items that you would be reporting as having value, including things like printing presses, lines of code, intellectual property, obligations from people who may or may not pay you in the future, cash in various countries’ currencies, and best guesses on things like the future prices of a commodity or the value of another company.

As an auditor, investor, or trading partner, you might want to drill down and try to test the assumptions that the company is making and see what would happen if those were incorrect at the time they were recorded, or turned out to be wrong sometime in the future. You might also want to understand how buying another company would change your own company based on the way your obligations and bets interacted with theirs. You could rack up millions of dollars in auditor fees to “get to the bottom” of any number of assumptions. The process would involve manually reviewing the legal contracts, and also the assumptions made in every cell of every spreadsheet. That’s because standard accounting is a very “lossy” process that reduces complex and context-dependant functions and transforms them into static numbers at every step. The underlying information is somewhere, but only exposed with a lot of manual digging.

The modern complex financial system is full of companies that have figured out ways to guess when investors and the companies themselves have made mistakes in their assumptions. These companies bet against a company with inaccurate pricing or take advantage of the gap in information to convert this into financial returns for themselves. When these mistakes are duplicated across the system, it can cause fluctuation amplification that also allows companies to make more money both as markets rise, as well as fall, if they can successfully predict those fluctuations. In fact, as long as the whole system doesn’t collapse, smart traders make more money on fluctuation than on stability.

Just like rodent exterminators aren’t excited about the idea of rodents being completely eliminated because they would no longer have jobs, those financial institutions that make money by “making the system more efficient and eliminating waste” don’t really want a stable system that isn’t wasteful.

Right now, the technology of the financial system is built on top of a way of thinking about money and value that was designed back when all we had were pen and paper, and when reducing the complexity of the web of dependencies and obligations was the only way to make the system functionally efficient. The way we reduce complexity is to use a common method of pricing, put elements into categories, and add them up. This just builds on 700­-year­-old building blocks, trying to make the system “better” by doing very sophisticated analysis of the patterns and information without addressing the underlying problem of a lossy and oversimplified view of the world: a view where everything of “value” should be as quickly as possible recorded as a number.

The standard idea of the “value” of things is a reductionist view of the world that is useful to scale the trading of commodities that are roughly of equal worth to a large set of people. But, in fact, most things have very different values to different people at different times, and I would argue that much–if not most–things of value can’t and probably shouldn’t be reduced to numbers on a spreadsheet. Financial “value” has a very specific meaning. A home clearly has “value” because someone can live in it and it’s useful. However, if no one wants to buy it and no one is buying similar homes on the market, you can’t set a price for it; it is illiquid and it is impossible to determine its “fair market value.” Some contracts and financial instruments are nonnegotiable, may not have a “fair market value,” and may even have no value to you if you needed money (or an apple) RIGHT NOW. Part of the confusion comes from the difficulty of describing legal and mathematical ideas in plain English, and the role of context and timing.

One example is exchange rates. My wife moved to Boston from Japan several years ago, but still converts prices into yen. She sometimes comments on how expensive something has gotten because the value of the yen has diminished. Because most of our earnings and spending are in dollars, I always have to remind her, the “value” in yen is irrelevant to her now, although not irrelevant to her mother, who is in Japan.

We have become accustomed to the notion that things have a “price,” and that “price” is equivalent to its “value.” But an email from you to me about a feeling that you had about our last conversation is probably valuable to me at a particular time and probably not valuable to most people. A single apple is worth a lot more to a hungry person than the owner of an apple orchard. Context is everything.

Can't Buy Me Love The Beatles

The economics notion of consumers making financial decisions to maximize “utility” as a kind of proxy for happiness is another example of how the notion of a universal system of “value” oversimplifies its complexity–so much so that the models that assume that humans are “economically rational” actors in a marketplace simply don’t work. The simplest version of this model would mean that the more money you had, the happier you would be, which Daniel Kahneman and Angus Deaton argue is true up to about $75,000 a year in annual income. [1]

Today, we have the technology and the computational power to create a system of accounts that could retain and deal with a lot of the complexity that the current system was designed to avoid. There is, for example, no reason that every entry in our books needs to be a number. Each cell could be an algorithmic representation of the obligations and dependencies that it represents. In fact, using machine learning, accounts could become sophisticated probabilistic models for what might happen depending on how things around them change. This would mean that the “value” of any system would change depending on who was asking, their location, and the time parameters.

Today, when a bank regulator conducts a stress test, it gives a bank a scenario–changes in the credit markets or the prices of certain things. The bank is then required to return a report on whether it would crash or remain solvent. This requires a lot of human labor to go through the accounts and run simulations. But what if the accounts were all algorithmic, and instead you could instantly run a program to provide the answer to the question? What if you had a learning model that could answer a more important question: “What sets of changes to the market WOULD make it crash, and why?” That’s really what we want to know. We want to know this not just for one bank, but the whole system of banks, investors, and everything that interacts.

When I’m buying something from a company–let’s say a credit default swap from your company, AIG–what I would want to know is whether, when the day comes to pay the obligation, in the unlikely chance that the AA mortgage-backed bonds that I was betting against defaulted, would your company be able to pay? Right now, there is no easy way to do this. However, what if all of the obligations and contracts, instead of being written on paper and recorded as numbers, were actually computable and “visible”? You’d immediately be able to see that, in fact, in the scenario in which you’d have to pay me, you’d actually have no money since you’d written similar contracts to so many people that you’d be broke. Right now, even the banks themselves can’t see this unless an internal investigator thinks to look for this ahead of time.

Rethinking the Fundamentals of Accounting

With cutting edge cryptography like zero-knowledge proofs and secure multiparty computation, there are ways we might be able to keep these accounts open to each other without compromising business and personal privacy. While computing every contract as a cell in a huge set of accounts, every time anyone asked a question it would exceed even today’s computing capacity. But with machine learning and the creation of models, we might be able to dampen, if not stabilize, the massive amplifications of fluctuations. These bubbles and collapses occur today, in part, because we are building our whole system on an oversimplified house of cards, with the handlers having an incentive to make them fragile and opaque in order to introduce inefficiencies they can exploit later to make money for themselves.

I think the current excitement about Bitcoin and distributed ledgers has created a great opportunity to take advantage of its flexible and reprogrammable nature, allowing us to rethink the fundamental system of accounts. I’m much more interested in this than in apps for banks, or even new ideas in finance,, which will address some of the symptoms without taking a shot at eliminating one of the root causes of the impossibly complex and outdated system that we’ve built on a 700 year old double-entry bookkeeping method–the very same system used by the Florentine merchants of the 13th century. It feels like we are using integers when we should be using imaginary numbers. Reinventing accounting should be more like discovering a new number theory than tweaking the algorithms, which is what I feel like we’ve been doing for the last several hundred years.

Citations:

[1]Kahneman, D. and Deaton, A. High Income Improves Evaluation of Life But Not Emotional Well-Being. Proceedings of the National Academy of Sciences.