Thursday, June 30, 2016

Algorithms stayed the chaos during Brexit storm: Can they help with auditor judgment?

The recent Brexit crisis hit the markets hard with the various stock indices plummeting and investors fleeing for the safe haven of gold, which went up "by $59.30, or 4.7 percent".

Amid this  chaos, some investment strategies fared well - thanks to the use of robots.

According to the WSJ article, "Who Made Money in the Brexit Chaos? Machines, Not Humans",  machines were immune to the fear, uncertainty and doubt that plagued markets (italics, highlight mine):

"This fund category, sometimes called commodity trading advisors, or CTAs, uses customized trading algorithms to spot market trends and place bets on futures and other derivatives. Most of the models didn’t factor in British election polls, bookmakers’ odds or the political-tea leaf reading that swayed other investors looking for an edge. In the weeks leading up to the Brexit vote, the trading models at many of these firms adopted a defensive pose. They favored high-quality government bonds, gold and safer currencies like the yen, while mostly avoiding riskier bets like oil and emerging markets.

That positioning paid off after Brexit caused the pound and more volatile assets to plunge as Thursday’s results came in. Société Générale’s CTA Index gained 1.5% on Friday. AQR Capital Management LLC, Fort and Welton Investments Partners LLC were among the big gainers... A key to CTAs’ success, their managers say, is that their models can tune out noise around market moving events—like an election or crucial economic data—that are important to investors but can be difficult to accurately forecast."

The article also quoted Lara Magnusen, portfolio strategist for Altegris’s main fund, who said (bold mine):

"Our models aren’t going to be affected by the same sentiments a human would be"

I thought that this was interesting as it illustrates how the machines can be seen as a way to provide an anchor when people are getting caught up in an emotional frenzy. Think of the implications for the world of audit and assurance, where professional judgement are made to determine what accounts, transactions, etc. are risky and should be tested. Imagine an audit algorithm that can be as an independent monitor that vets judgments of the audit professional - in a "race with a machine" scenario (for more on this idea see the Ted Talk below with MIT professor Eric Brynjolfsson). This could potentially improve auditor judgment, stakeholder confidence and audit quality.

Initially, I think this would be a way for audit firms to reduce the level of uncertainty associated with reviews from the PCAOB, CPAB and their equivalents in other jurisdictions. This would especially be the case if such audit oversight bodies would "bless" such algorithms and be able to ensure that the firms applied such judgment consistently, e.g. by having access to the "audit logs" produce by such programs.

The next - and more controversial step - would be to argue that independence rules can be relaxed in light of such automated oversight. To be honest I think there's a low likelihood of such an idea making traction with regulators in the near future, given that Europe has sought to require mandatory rotations of auditing firms. But it is something that should at least be contemplated, especially when automation becomes commonplace and attitudes may change towards how algorithms can play nicely with humans.

Tuesday, June 28, 2016

GoldmanSachs on Blockchain: Insights into Audit & Accounting Automation

As noted in this Business Insider article, Goldman Sachs (GS) published a report on blockchain that identifies a number of scenarios where the technology can save billions. The BI article extracted the following use cases from the 88 page report:
  • Better authentication of individuals partaking in the sharing economy: Leveraging the "smart identity" functionality of the blockchain, peer-to-peer sharing businesses sites (e.g. Airbnb) can give both the customer (e.g. the renter) and the supplier (e.g. the home owner) greater assurance that the customer is really who they say they are. The GS report also links the identity to smart contracts that facilitates automated performance based payments 
  • Accounting system for renewable energy power generation: Where individual homeowners are generating wind or solar power, the blockchain can be the natural accounting system to manage the "debits and credits" transferred back and forth between the energy producer and the network. It also enables payment transfers as well. 
  • Reducing back end administration for title insurance: The actual GS report notes how the vast majority of the cost associated with title insurance can be reduced by about 30% using blockchain to manage the underlying property records. Other interesting notes is that they attribute part of the decline to improved actuarial risk calculations due to "greater historical transparency". 
  • Improving accuracy and timeliness of trading various securities: The financial services industry usage of the blockchain is quite straightforward - replace the chaotic world of spreadsheet accounting with the streamlined world of blockchain - it is a database technology after all. NASDAQ use of Linq was featured in this DUPress article and can also be found here. The GS report goes into much more granular detail as to the different scenarios on how the back-end system can be improved resulting in less verification issues and improved trading times.  
  • Better authentication of customers aka KYC (Know-Your-Customer): As noted in the BI article, "Like with the Airbnb example, Goldman envisions identity data stored on a blockchain that could help finance firms easily and quickly check new customers as part of "know your customer" regulation — a bit like a digital passport."
I went through the long report and extracted the following accounting automation insights:
  • Blockchains can reduce spreadsheet funk. Sharing spreadsheets has become the norm in the financial world for being a flexible way to send quantitative information along with context and some formulas here and there; it's how we auditors often get data from the client when asking for a breakdown of an account we are looking into as a part of the audit. It's also error prone. Blockchain, in a sense, is a "napster-esque" way of sharing financial information that ensures a common data structure between the sender and the receiver thereby eliminating the manual verification/handling of spreadsheets (see pages 3 and 10).
  • Blockchain can enhance "assurance", where it's not feasible for auditors to do so. On page 16 the report discusses the role of smart identities in assisting the sharing economy. It talks about how required digitally signed user reviews will have greater data integrity as it could reduce the risk of self-inflated reviews. Although people rely on such reviews to buy books, rent hotels, etc., there is risk of fraud where the seller will inflate reviews or pay people to do so. However, with a blockchain enabled smart identity there's a higher level of assurance that the end-user can place on the reviews as it harder for the site owner to fake the reviews due to fact the review is digitally signed. Of course no audit firm would have audited such reviews. But I think that's the point: the blockchain technology fills an an assurance need that auditors couldn't, simply because the delivery of such a service wouldn't be profitable. 
  • Blockchain automation will reduce the need for back-end clerical staff (aka accountants): When looking at the application of blockchain to the title insurance industry (see pages 33-39), it notes how 75% of the industry premiums relate to headcount costs. GS puts the reduction in clerical staff by 30% and a 20% reduction in variable expenses (e.g. commissions, marketing, etc.). Blockchain - without AI enhancements - will automate accounting work as part of the automating knowledge work trend. This is of course more clerical tasks, but blockchain will likely result in less headcount within the finance department. 
  • Role for third party assurance reports in a permissioned blockchain: The consensus mechanism in the permissioned blockchain is quite different than it's public counterpart, which relies on the proof of work (POW). (See the Khan Academy video, below for the POW and the blockchain section in this post) This is not the case for permissioned system which require the consortium who set-up the blockchain to determine how they will work with each other. This could require auditors to provide assurance over the implementation of blockchain similar to what the SOC report does for cloud computing companies. The report discusses how (see page 29) on how the blockchain will enable "Smart Grid Blockchains" to essentially acts as the record keeping and payment system of energy exchanged by the household owner who has windmills, solar panels to the power grid. But how do we ensure that this being calculated properly? Well, that's where the Processing Integrity Principle of the SOC3 assurance report comes in. It could provide assurance that the blockchain-accounting-payment system is processing the data in complete, accurate and timely manner.  
  • Greater visibility, means greater opportunity for audit analytics. One area of cost savings associated with a blockchain enabled title insurance industry is that actuaries will be better able to assess risk  because of "greater historical transparency and immutability into the property registration system" (see page 38). Consequently, where a material amount of transactions are on a blockchain auditors will have (1) easier access to the data (not a trivial matter by any means!) and (2) can run better analytics to identify irregular transactions and (3) enable better ways to assess estimates. 
  • Value versus hourly billing: In a number of the use cases identified (e.g. title insurance, settling equities, KYC; see pages 38, 51, 75) noted how the gains (read: headcount reduction) from blockchain enabled automation are expected to be passed on to the customer. Why is this relevant to audit? Audit firms could be expected to hand over automation windfalls to the client and further reduce fees. On the one hand, the more automated the audit, the potentially less fees that audit will capture. On the other hand, regulators may want the auditors to do more with the budget that has been freed up. So the revenue, profitability of highly automated audits will depend on how the regulators re-draw scope in light of such advances. 
Despite having the reputation as a great-vampire-squid, the GS report is quite useful for those working in the blockchain space in identifying the potential for this exponential technology.

Tuesday, June 14, 2016

AI-as-a-Screenplay Writer: Computer overlords strike again?

Normally when I discuss artificial intelligence on the impact of work, it's in the context of the automation of accounting and auditing work (A3W). However, a story that's been circulating for past few days in my Google alerts is the story of the movie Sunspring. Unlike most movies it's not written by human being, but rather it is written by a computer.

Ars Technica, who hosted the online debut of the movie, noted that the script was "authored by a recurrent neural network called long short-term memory, or LSTM for short. At least, that's what we'd call it. The AI named itself Benjamin."
The movie is really odd to put it nicely. However, it does comes across as one of those art movies that (also) don't make any sense. The song in the movie is also generated by the machine.

However, this is not the first time that algorithms have been trained to be artists. Chris Steiner in his book "Automate This: How Algorithms Came to Rule Our World" notes that Emmy,  an algorithm, "produced orchestral pieces so impressive that some music scholars failed to identify them as the work of a machine". In a piece he authored for the Wall Street Journal he notes "analyzing only the script, an algorithm from Epagogix, a risk-management firm that caters to the entertainment industry, predicts box office grosses. Epagogix broke into the business when a major studio allowed the firm to analyze script data for nine yet-to-be released films. In six of the nine cases, its predictions were spot-on. Algorithms have since become an essential tool in Hollywood."

If the chaotic world of creative works can be automated by algorithms, then I think the predictable, routine world of debits and credits can't be too far behind.

Monday, June 13, 2016

Can accounting errors ruin your life? JohnOliver explains how they can.

In this episode Last Week Tonight, John Oliver explores the world of debt buying:

The segment received wide publicity as he tried to out do Oprah by conducting the biggest giveaway on television - he bought $15 million worth of medical debt and forgave it. This article on Fortune does a good job of summarizing the show:
  • US households owe $12 trillion in debt of which $436 billion is 90+ days past due. 
  • Companies who discharge the debt sell it for pennies on the dollar to a growing number of companies that specializes in debt buying.  
  • One company, Encore Capital, notes that in 1 in 5 Americans owes or has owed them money. 
  • Debt that's been paid "come back to life", which is affectionately known as Zombie debt.
There was some controversy, however, about who he worked with to write-off the debt (they noted their grievances here, to which John responded here) and the value of the debt. On the latter count, is it really fair to criticize an act of charity that improved the lives of approximately 9,000 people?  

Nothing good happens in Excel. 
But the segment which is most relevant to us is when he starts talking about how the information is actually sold.  It is sold on spreadsheets. Oliver gets quite dramatic as he shares his his phobia of Excel and notes how "nothing good happens in Excel". He also explains that the spreadsheets are sold "as is"; meaning that the seller does not guaranty accuracy of the information related to the debt contracts being sold.

And that's where the jokes stops.

In the segment, he has footage from interviews with Jake Halpern, who wrote "Bad Paper: Chasing Debt from Wall Street to the Underworld". The book follows the life of a debt buyer of Aaron Siegel, who is born to a rich family in Buffalo, New York. He takes an array of characters, including his Brandon, who is an ex-con who does that gritty part of work of finding the debt, ensuring its good and collecting on it.

What caught my attention as I was going through book, is that it gave a bit more detail to what John Oliver mentioned about the banks selling the paper "as is". Halpern notes on page 58 of his book (see below for the link to the book), that when Washington Mutual sold Joanna and Theresa's debt to Aaron, the credits awarded against their accounts that were not reflected in the spreadsheet that was given to the debt buyer.

And that's how accounting errors can ruin lives.

When you read the life stories of these two ladies it's heart wrenching to think that a few lines on an Excel spreadsheet could have a detrimental impact on their lives. Some would cynically say this is over dramatic and try to find reason to blame Joanna and Theresa falling into this problem. But I don't think that's fair. When you read the lives of these people, it's clear that they were affected by factors beyond their control. It's really this broken system of debt collection that is responsible for them failing to get the debt relief that they were owed.

The way accounting systems and spreadsheets are designed and operated can have real impact on real people. As an accountant myself, I often wondered what value is accounting in the grand scheme of things. But as Halpern's story illustrates the accountants, bookkeepers, etc. had a real impact on the livesof these two women.

No one is saying that accountants have the same impact on the lives of people the way a cancer specialist does. But at the same time a few a lines on Excel spreadsheet could be the difference between perpetual anxiety and a good nights sleep.