Saturday, August 31, 2019

Why the discrepancy?

Vitalik Buterin had a thought-provoking tweet a few days back about interest rates.
Today's post explores what goes into determining interest rates, not blockchain stuff. So for those who don't follow the blockchain world, let me get you up to speed by decoding some of the technical-ese in Buterin's tweet.

DAI is a version of the U.S. dollar. There are many versions of the dollar. The Fed issues both a paper and an electronic version, Wells Fargo issues its own account-based version, and PayPal does too. But whereas Wells Fargo and PayPal dollars are digital entries in company databases, and Fed paper dollars are circulating bearer notes, DAI is encoded on the Ethereum blockchain.

Buterin points out that DAI owners can lend out their U.S. dollar lookalikes on Compound, a lending protocol based on Ethereum, for 11.5%. That's a fabulous interest rate, especially when traditional dollar owner can only lend their dollars out to the government—the U.S. Treasury—at a rate of 1.5%.

Why this difference, asks Buterin?

Interest rates are a lot of fun to puzzle through. I had to think this one over for a bit—so let's slowly work through some of the factors at play.

Let's begin by flipping Buterin's question around. When the U.S. Treasury borrows from the public, the bonds it issues are promises to pay back regular dollars (i.e. Federal Reserve dollars). But what if the U.S. Treasury decided to borrow DAI by issuing bonds promising to repay in DAI? What would the interest rate on these Treasury DAI bonds be? Would it be 11.5% or 1.5%? Perhaps somewhere in between?

Credit risk

First, there's the question of credit risk. The U.S. Treasury is a very reliable debtor. It won't welch. If it issues both types of bonds, it'll be just as likely to repay its DAI bond as it will its regular dollar bond. Since the market already requires 1.5% from the Treasury to compensate it for credit risk (and a few other risks), the Treasury's DAI bonds should probably yield 1.5% too. (I'll modify this later as I add some more layers).

Now let's look at Compound. A DAI loan made on Compound (for simplicity let's just call it a Compound DAI bond) is surely much riskier than our hypothetical Treasury DAI bond. Compound is a blockchain experiment. It could malfunction due to buggy code. Maybe every single Compound borrower goes bust. To compensate for this risk, a prospective bond buyer will require a higher return from Compound DAI bonds than they will U.S. Treasury DAI bonds.

So Compound credit risk (Buterin's third option) probably explains a big chunk of the huge gap between the 11.5% interest rate on Compound DAI bonds and our hypothetical 1.5% interest rate on the U.S. Treasury's DAI bonds. But not all of it.

Collapse risk

Buterin mentions a second risk: the chance that DAI, the entity that creates blockchain dollars, collapses. Like Compound, DAI is a new monetary experiment. The code could be buggy. It might get hacked. By comparison, conventional dollar issuers—say Wells Fargo or PayPal—are far less likely to malfunction.

How does DAI collapse risk get built into the price of a hypothetical Treasury DAI bonds

The average market participant (I'm not talking about crypto fans here, but large & smart institutional actors) should be genuinely worried about purchasing a Treasury DAI bond—not so much because the Treasury is unlikely to pay it back—but because the DAI tokens that the Treasury ends up repaying could, in the even of DAI breaking, be worth 99% less than their original value. Average bond buyers will expect some compensation for bearing this risk. How much? Say 5.5% (I'm just guessing here).

Earlier I said that a Treasury DAI bond would yield 1.5%. But if we add 5.5% worth of failure risk to 1.5% in basic risk, a Treasury DAI bond should yield 7.0% before the average investor is going to hold it.

Now let's go back and look at a Compound DAI bond. As Buterin pointed out, they yield 11.5%, which is much higher than the 7.0% yield on our hypothetical Treasury DAI bond. We've already assumed that DAI collapse risk works out to 5.5%. If we subtract collapse risk from a Compound DAI bond's 11.5% yield, the remaining 6% is accounted for by risks such as Compound failing (11.5% - 5.5%). Put differently, investors in Compound DAI bonds will require 5.5% and 6.0% to compensate for collapse risk and credit risk respectively, for a total of 11.5%. Again, these are hypothetical numbers. But they help us puzzle things out.

Two different blockchain dollars: USDC vs DAI

Interestingly, Compound doesn't just facilitate DAI loans. It also expedites loans in another blockchain dollar, USDC. We'll refer to these as Compound USDC bonds. As Buterin points out later on in the thread, the rate on Compound USDC bonds is 6.5%, quite a bit lower than Compound DAI bonds.

What might explain this discrepancy?

Not credit risk, since in both instances the same creditor—Compound—is responsible for creating the bonds. Which leaves varying levels of collapse risk as an explanation. USDC is a regulated stablecoin (i.e. it has the government's approval). DAI isn't. And USDC has genuine U.S. dollars backing it, whereas DAI is backed by highly volatile cryptocurrencies. So the odds of USDC collapsing are surely lower than DAI.

How much interest do USDC bond holders require to compensate them for collapse risk? Assuming that Compound's risk of failing is worth 5.5% of interest (as we already claimed), that leaves just 1% attributable to the risk of USDC failing (6.5%-5.5%). Put differently, investors in Compound USDC bonds will require 5.5% and 1.0% to compensate for credit risk and collapse risk respectively, for a total of 6.5%.

Oddly, the yield on a Compound USCD bond is less than the hypothetical yield on our safe Treasury DAI bond (6.5% vs 7.0%). Why is that? Even though Compound is riskier to lend to than the Treasury, a DAI-linked return is riskier than a USDC return. Another way to think about this is that if the Treasury were to also issue USDC bonds, those bond would only yield 2.5%. To account for credit (and other) risks investors would require a base 1.5% with an extra 1.0% on top for the risk of USDC breaking.

The convenience yield

Let's bring in one last layer. Something called the convenience yield is lurking behind this.

When you lend me some tokens, you need to be compensated for more than just credit risk i.e. the risk that I won't pay back the tokens. You are also doing without the convenience of these tokens for a period of time. The replacement, my IOU, won't be very handy. For instance, the convenience of a dollar bill can be though of as the ability to mobilize it whenever you need to meet some pressing need. But if you've lent a $100 bill to me then you've given up all that bill's usefulness. Instead, you're stuck with my awkward $100 IOU. You need some compensation for this. (Unconvinced? Head over to Steve Randy Waldman's classic ode to the convenience yield).

So when we break down the components of the interest rate on DAI bonds, there must be some compensation required for forgoing the convenience of DAI, its convenience yield. Earlier I attributed the big gap between rates on Compound DAI and USDC bonds to varying odds of each scheme failing. However, the gap could also be explained by varying convenience yields. If the convenience yield of a DAI token is higher than that of a USDC token, we'd expect an issuer of a DAI bond to pay a higher rate than on a USDC bond, in order to compensate DAI holders for giving up on those superior conveniences. 

If DAI's convenience yield is higher than USDC's, what might explain this gap? DAI is completely decentralized and can't be monitored. USDC isn't. It is less censorship-resistant than DAI. So perhaps USDC just isn't as handy to have around.

So some of the 11.5% rate on Compound DAI bonds—say 2%—may be due to the convenience yield forgone on lent DAI. If DAI had the same features as USDC, and thus had a lower convenience yield, a Compound DAI bond might only yield 9.5% (11.5% - 2.0%). If so, the discrepancy between the Compound DAI and USDC bonds—9.5% vs 6.5%—wouldn't be as extreme.

Summing up, let's revisit Buterin's tweet:
If my line of thinking is right, the discrepancy is accounted for a messy mix of the higher risk of lending to Compound (3), the danger of DAI cracking (2), and whatever convenience yield one forgoes when one no longer has DAI on hand (4-other). And of course, Buterin's first option is right too. I'm assuming that people are rational and can easily buy and sell various assets. But the sorts of large institutional players who set market prices may not be operating in crypto markets.

Wednesday, August 21, 2019

Starbucks, monetary superpower



I recently spent some time on Twitter discussing the monetary wonders of Starbucks. In this post I'll bring a bunch of tweets together into a single blog post.

I don't go to Starbucks very often, so I only recently learnt that the company has succeeded in getting many of its customers to stop using cash and debit/credit cards to buy coffee. Instead, they are using  Starbucks's own payments option:
Starbucks has around $1.6 billion in stored value card liabilities outstanding. This represents the sum of all physical gift cards held in customer's wallets as well as the digital value of electronic balances held in the Starbucks Mobile App.* It amounts to ~6% of all of the company's liabilities.

This is a pretty incredible number. Stored value card liabilities are the money that you, oh loyal Starbucks customer, use to buy coffee. What you might not realize is that these balances  simultaneously function as a loan to Starbucks. Starbucks doesn't pay any interest on balances held in the Starbucks app or gift cards. You, the loyal customer, are providing the company with free debt.

Starbucks isn't the only firm to get free lending from its customers. So does PayPal. That's right, customers who hold PayPal balances are effectively acting as PayPal's creditors. Customer loans to PayPal currently amount to over $20 billion. Like Starbucks, PayPal doesn't pay its customers a shred of interest. But Starbucks's gig is way better than PayPal's. PayPal is required to store customer's funds in a segregated account at a bank, or invest them in government bonds (see tweet below). So unfortunately for PayPal, it earns a paltry amount of interest on the funds that customers have lent it.

Starbucks, on the other hand, doesn't have to keep customer funds in a low yielding segregated account or government bonds. Why is that? PayPal allows people to cash-out of PayPal dollars into regular dollars, so for regulatory purposes it must keep an adequate reserve on hand to facilitate redemptions. But the only way to cash out of Starbucks balances is to buy a coffee--a promise that Starbucks can always keep! And so Starbucks can immediately put its customer loans to work in higher-yielding opportunities like funding its operations and expansion.

In addition to borrowing from its customers, Starbucks also borrows from professional investors. Here's a list Starbucks's long-term debt:


Starbucks is paying an interest to bond and note-holders that ranges as low as 0.46% (on its yen notes maturing in 2024) to 4.5% (on its 2048 notes). You can see why borrowing from customers in the form of stored value card liabilities is the better option. By expanding its borrowing from its non-professional lenders and using the proceeds to cancel its debts to professional lenders, Starbucks can make an immediate profit.

But there's more. As I pointed out in the following tweet, don't forget breakage. Bond and note holders are pros. They don't forget about debts. But customers aren't so exact. They are sloppy, or busy, or forgetful, which means that many gift cards and balances will go unspent:

Each year Starbucks recognizes that a portion of its stored value liabilities will be permanently lost. This is known as breakage. Starbucks recognizes this amount as profit. In 2018 the company recognized $155 million in breakage, around 10% of all stored value balances. Wow! Starbucks already pays just 0% on its debts to customers, but add in breakage and that equates to a roughly -10% interest rate!

On Twitter, Wayne points out to me that I need to add back the impact of Starbucks rewards. App users receive stars on each purchase which can be saved up for free coffee. This functions as a form of implicit interest that Starbucks pays to its customers.

That's a good point. But if were going to bring rewards into our calculation, then there are other non-pecuniary flows that need to be added in too. Keep in mind that each payment made through the Starbucks app is a payment that isn't made by credit card. Since each credit card payment will cost Starbucks 1-2% in interchange fees paid to the card networks and banks, the company saves a lot of money by guiding customers to its payments app. As for Wayne, while he may earn an implicit interest return in the form of Starbucks points, by forgoing a card payment he's giving up on the associated cash-back or airline points.

Another flow that needs to be accounted for is data. By capturing the customer's wallet, Starbucks is getting loads of free but valuable personal information that would otherwise be lost, or for which it would have to pay. Any customer who pays with cash forgoes rewards, but at least they get to retain their information. 

Adding all of this up, (0% interest + breakage - rewards + interchange savings + customer information), Starbucks's stored value liabilities are a terrific liability to have.

More generally, I think this calculation demonstrates how providing financial services to a retail customer base is a great business. Retail customers don't seem to be too fussy about the return they get. And they are busy and distracted and sloppy and forgetful. Take central banking, for instance, which serves a retail clientele. People are pretty happy to hold banknotes that pay 0%. But you never see businesses or professional investors hoarding banknotes. They quickly return the cash they take in during the course of the day to their bank so that they can harvest interest. Commercial banking is also a good example. Like Starbucks, banks are able to borrow from their retail customers at a measly rate approaching 0%. But professionals who lend to banks by purchasing their bonds require a much higher rate. To top it off, retail customers unnecessarily sign up for high-fee products and avoid changing banks when there is a cheaper option.  

Why doesn't every retail chain try get into this game? By borrowing as much as they can from the non-professional public, they'd steal plenty of profitable business from central banks and retail banks. Well they do. Gift cards are a big business. And if you think about it, retailers are perfect candidates for providing monetary services to the masses. Like banks, they already have a network of physical stores. But none of them have been successful at it as Starbucks. Walmart is much bigger than Starbucks, for instance, but it has just as many gift card balances outstanding:

Perhaps Starbucks's success has to do with the regularity and homogeneity of Starbucks purchases? And so customers are willing to preload a dedicated account? I'm not sure.

In any case, there are probably a few Starbucks executives who'd love to grow the amount of negative yielding liabilities that the firm issues. Why stop at $1.6 billion in stored value liabilities? Why not grow the program to $5 billion, $10 billion, or $100 billion? It would be a terrific business line to get into.

The problem here is that Starbucks only sells coffee. Coffee is great, but the demand for dollars that are only useful for buying coffee will always be limited. To really grow the amount of stored value liabilities it issues, Starbucks would have to increase the usefulness of Starbucks dollars. One way to do this would be to open up the Starbucks app up to other stores. If consumers could also buy Big Macs with the balances on Starbucks App, this would increase the demand for Starbucks balances. To secure McDonald's cooperation, Starbucks would have to share the savings, breakage, and data. Maybe companies like Home Depot and Costco would join the Starbucks-McDonald's alliance. (And other chains, say Kroger and Burger King, might join the competing Walmart Pay alliance).

Sure, each of these companies could simply pursue their own independent stored-value liability programs. But wouldn't an alliance be better? From the customer's perspective, balances held in a single payments app that can be spent at Starbucks, McDonald's, Home Hardware, or Costco would be far more useful then dollars held in four separate and walled-off apps. And so collectively these stores should be able to get the public to hold more stored value card liabilities than they could individually. Which means more breakage, free loans, and data for everyone (and less for the banks, card networks, and central banks).

Who knows if it would be successful. And it might not even be possible from a regulatory perspective. But it would be tempting, no? In a world where most debtors have to pay interest, being  a debtor who earns interest is pretty hard to beat.



*I believe that current portion of deferred revenue is equal to around $174 million. This comes courtesy of the current portion of an up front royalty payment from Nestlé. So the stored value card liability is actually closer to $1.46 billion. Still pretty high. 

Monday, August 5, 2019

Stigmatized money


Some payments systems are so awkward they scare away the average user. The only people with the patience to stick around must have a motivation for doing so. These include ideologues with an ax to grind, hobbyists who happily embrace complicated features, and criminals/weirdos who are shut out of everything else.

Here are a few examples of awkward payments systems:

-Local Exchange Trading Systems, or LETS
-Bitcoin/Dogecoin
-Labor notes
-Stamp scrip

When usage of a payments system is confined to a narrow group of like-minded individuals, this may stigmatize these systems, scaring away mainstream users. Stigmatization only compounds the initial awkwardness. After all, if fewer venues accept the stigmatized payments option then it becomes harder for the small band of users to make purchases. A vicious circle has been created. Initial awkwardness leads to stigma which leads to more awkwardness etc.

While this vicious circle is the death knell for a payments system, it is less of a problem for other products. You can make a decent living by targeting a small niche of consumers, say communists who eat vegan food. Every big city needs a communist vegan restaurant. But a payments network is only as good as the size of the payments pathways that it facilitates. A payments entrepreneur won't get very far by building a platform that only allows communist vegans to pay other communist vegans. 

How to evade the awkwardness-stigma spiral? What is needed is a frictionless, non-awkward payments system. With little to learn, everyone—not just nerds and those with an ax to grind—can quickly start using it. Think M-Pesa or Visa or Octopus.

Even the haters will get onboarded. Gold bugs and bitcoiners may rail against banks and fiat money. But because everyone else is using these relatively simple systems, the bugs and the bits have no choice but to go along. By bringing the vast hoard of normies on board along with the weirdos, these systems avoid all connotation. They are safe for broad consumption. No stigma can attach to them. And so the vicious awkwardness-stigma circle I described gets sidestepped.

I'd argue that LETS are an example of a system that suffer from the awkwardness-stigma spiral. LETS are a pain to use. This article on the famous Comox LETS and its founder Michael Linton, an earlier proponent of LETS, explains some of the problems. And so the only people who use LETS will be those willing to put up with the awkwardness: folks who self-identify as leftist with anarchic views. Those who don't share those views might feel weird joining a LETS. So LETS remain small and fragile, or as Linton says, they're like "sandcastles on the beach."

You see the same awkwardness-stigma cycle at play in bitcoin. Bitcoin is an awkward payments medium. The stuff is so volatile that retailers don't like to accept it. Risk averse consumers don't want to hold it. So only a subset of the population will ever feel comfortable using bitcoins for payments.

This subset has developed its own norms and codes. Bitcoin steak dinners are a good example. A large group of predominately male bitcoiners will get together to eat meat while avoiding vegetables, then broadcast it on Twitter:

I'm sure it's a lot of fun. But these sorts of traditions will inevitably be perceived as weird by the majority. And so the majority will go out of their way to avoid bitcoins for fear of being tarred as an oddball. Other niche groups who can't sympathize with male carnivores, say lesbian vegans, will avoid bitcoin payments on principal. This stigma cuts down on the potential pool of bitcoin payees, which only makes bitcoin more annoying to use.

Mastercarders don't have their own set of traditions. For instance, you won't see Mastercard users setting up meetups to eat organic food and talk about the latest development in tokenization technology. The Mastercard/Visa user-bases are devoid of culture and character. This lack of a class consciousness is one of the features that makes them such effective  payments networks. Systems without norms and traditions never face the risk of falling into the stigmatization hole.

Facebook's Libra has attracted plenty of attention over the last few months. But Libra risk encountering the same awkwardness-stigma cycle as Bitcoin and LETS. Unlike other social media-based  payments tools (Wechat, Kakao, Line etc), Libra's architects have chosen to create a new unit of account rather than marrying Libra tokens to existing units like the dollar or euro.

But as I suggested in a previous post, it's a pain to learn a new unit of account, just like it's a hassle to learn a new language. So only a certain type of motivated person will bother using Libra, just like only motivated people—language nerds—learn Esperanto. This weirdo factor could stigmatize the system. Hey, look at those Libra-using elitists! What snobs! By crowding out normies (and the massive number of potential payment pathways they bring to the table) Libra runs the risk of never self-actualizing as a payments system. Better to take the safe and boring route of linking Libra to dollars and yen and whatnot. 

The vicious awkwardness-stigma cycle has already started to hit cash. In places like Sweden, cash is being stigmatized. When the middle and upper class are convinced that only the poor and criminals use coins and banknotes, many of them will go out of their way to avoid using cash. Cash becomes "grungy and unsexy," as Brett Scott puts it. Unfortunately, an ever narrower base of cash users will only make the stuff more expensive for retailers to handle, leading to a rise in cashless stores (especially ones that cater to the rich), leading to more awkwardness and stigma, leading to less users, etc.

David Birch has an interesting parable from William Gibson's Count Zero that illustrates what happens when this stigmatization is brought to its logical conclusion. Basically, cash is still around in Gibson's imagined future, but it has disappeared from "polite society". And so the story's protagonist, Bobby Newmark, describes it as unspendable:

If cash is to avoid a Gibsonian future, it needs to be de-stigmatized. But this requires that it re-attracts many of the people and businesses that have deserted it because it is no longer convenient. In a post  at the Sound Money Project (and earlier on this blog) I suggested paying interest on cash. Thus individuals and businesses would be compensated for the relative inconvenience of note storage and handling. And with a wider range of people using notes, any stigma that they have attracted would dissipate.

Stigma is dangerous for a payments system. A system will become stigmatized if it attracts a clique rather than a broad group of users. Cliques kill a payments system since they suppress the system's connectiveness. To avoid the potential for clique-ization, systems should try to be as easy to use and accessible as possible.