Separating Hyperplanes

Yesterday at 2pm the CBO released a big report with estimates for lots of different budget options. I reviewed the health-related ones.

But first, a disclaimer: there's less health policy in the CBO releases than you might expect. You won't, for example, find much analysis on how federal health programs affect mortality, or estimates of the cost-effectiveness of public health insurance, nor even ideas for reducing the cost of delivering care to medicaid enrollees. So, don't think of this CBO document as being about policy. More realistically, you should think of it as a list of possible "pay fors" that the next congress can fish to offset the cost of new deficit-increasing policies republicans would like to enact.

Relatedly, there's very little in the CBO estimates that represents genuine "savings" to the government—most of the deficit reduction comes from proposed tax hikes. Sometimes this is explicit: one of the largest budget items in there is the proposal to tax 50 percent of the value of employer-sponsored health insurance plans, which would raise $430 billion over the first decade. Other times, this is implicit, such as the proposals to raise deductibles for medicare, VHA, and trihealth beneficiaries—after all there's little practical difference, for example, in raising the Medicare part A premium versus levying a tax on social security income. Other proposals include a mix of policy design, spending cuts, and tax hikes, such as the proposals to hike copays and coinsurance to medicare, which in addition to producing revenue, should also reduce hospital stays and other healthcare utilization by beneficiaries. So, there are some actual policy ideas in there, and a couple of them are pretty decent.

Federal Employees

Here's a pretty typical proposal in the genra: a plan to replace the Federal Employee Health Benefit with a voucher tied to inflation. There's very little policy here, as federal employees already shop for private insurance plans in the FEHB system. Rather, this is just a reduction in pay to federal employees: by tying the value of the health benefit to inflation (there's no good reason to suppose health costs won't grow faster than inflation), this proposal reduces federal payrolls by about $58 billion over ten years. This could be equivalently phrased as a revenue enhancement: a tax on federal employees. This is why I say we should think of these CBO options as a pond where Congress can go fish for money, not serious policy analysis.

CBO estimates that applying the same medigap restrictions (see below) to the military health benefit, Tricare, would save $27 billion. Most of this, about three fifths, savings comes from reduced benefits to beneficiaries (cost shifted from government to beneficiaries), while the remaining two fifths comes from reduced healthcare utilization. Also like the Medicare reform, this would cap out-of-pocket expenses at $4,125, making it a decent policy trade. Still though, "soldiers are overpaid" is not complaint I hear often—why not refund that $27 billion to them in the form of lower premiums?

You can save $29 billion by repealing VA health benefits for 2 million veterans. Half of those veterans will end up on other forms of public insurance such as medicaid or else receiving federal subsidies for Obamacare, while the other half will ¯\_(ツ)_/¯. These aren't veterans with disabilities, so ideally they'd be able to get health insurance through their post-military employment in private businesses.

Here's $18 billion if you raise premiums, deductibles, co-payments for veterans. There's no breakdown of how much is revenue/out-of-pocket versus how much is reduction in healthcare utilization, though we can safely assume most of it is the former. Hiking premiums makes Tricare less attractive relative to private insurance options, which I guess is desirable if you prefer they seek private insurance instead of a federal retirement benefit. On the other hand, this undermines the goal of keeping Tricare as a safety net for veterans who cannot afford private insurance.


Here's a proposal to modernize Medicare by capping out-of-pocket payments at $7,500 while raising the part A coinsurance to match part B's 20 percent, and adding a $250/day copayment for the first 5 days of hospital stays. That contrasts sharply with the current weak design of Medicare that not only doesn't cap out of pocket costs, but actually increases them for the sickest patients. The strengths of this policy design, however, are undermined by the need to raise revenue—rather than using the savings from decreased moral hazard and the revenue from co-pays and coinsurance to reduce premiums, the plan hikes premiums, imposing a $19 billion tax on seniors. But why?

Additional Medigap option: There's also an alternative in the above proposal to produce even more savings by restricting medigap policies so that can't pay for all of a senior's medicare copays and coinsurance. Under the restrictions, medigap could not cover the first $750 of patients' out-of-pocket medicare expenses, and no more than 50 percent of expenses over $750 but under $6,750, and this saves $45 billion over six years. Overall this is a dumb plan because it does not include other types of supplemental insurance such as employer benefits, but the estimates they have are noteworthy: in the first year, 2020, there will be 40 million fee-for-service enrollees and 20 percent of them will have medigap, and the CBO estimates restricting those medigap plans will save $4.9 billion, for a total of $6,125 in savings per beneficiary. Note that average Medicare spending is only $10,000 per beneficiary. CBO says they used similar behavioral estimates as RAND but a 60 percent reduction in healthcare utilization is quite a ways away above RAND estimates. It also doesn't seem consistent with estimates they used for the out-of-pocket cap above.

Here's a $330 billion tax hike on seniors, in the form of increases in premiums for Medicare part B and part D.

Medicare compensates healthcare providers for 65 percent of certain medical debts that patients can't pay. Reducing that to 45 percent saves $15 billion, and reducing it further saves $31 billion. To an extent, this is just a plan to tax healthcare providers, though I suppose they'll make up a portion of the loss by not treating poor patients, and squeezing those who struggle to pay even harder.

Here's medicare part D rebate program that looks to me a lot like a $145 billion tax on drug companies.


It turns out there's a loophole in federal medicaid funding to states that allows states to rope in more than the law intends. To simplify a bit, the federal government pays a percentage of medicaid expenses, and states levy special taxes on medicaid providers to artificially drive up measured medicaid spending, giving them more federal funding despite the fact that their actual medicaid expenses did not go up. This proposal narrows, but does not eliminate, this loophole, saving up to $40 billion over ten years.

But if that's not enough, it turns out that cutting medicaid by $374 billion saves almost $374 billion. Who knew? There's even less to this policy proposal than it sounds. There's no policy ideas about ways to reduce the cost of providing medicaid, but rather just a plan to reduce the pot of medicaid funds available to states. It does this specifically by imposing a per-beneficiary cap with annual increases pegged to CPI-U inflation rate, plus one percentage point, but the details hardly matter. The key point is that funding won't keep pace with healthcare costs, and this saves the federal government boatloads of money.

Graduate Medical Education

Now here's a plan to cut federal subsidies for Graduate Medical Education by $32 billion. Which, I guess I'm ok with? I've not found a convincing argument for why the government subsidizes this in the first place.

Medical Research

Here's a plan to reduce medical research by $9 billion. The premise of the plan is baloney—there's no analysis of the cost-effectiveness of the research, just a bogus argument that it doesn't belong in the Department of Defense. Ok, fine, let's move the money to NIH. Next?

Cigarette Tax

Ok, here's some real policy wonkery: a $0.50 per pack tax hike on cigarettes would raise $34 billion in revenue, and over the first decade save an additional billion in reduced public health expenses. After 10 years, as those healthier people live longer that $1 billion savings is likely to turn slightly negative. But besides, the reason to do this is the benefit to public health, not the revenue or savings.

Capping Malpractice Damages

I'm a little confused on how this proposal works, but it looks like capping medical malpractice suits at $750,000 (up to $500,000 punitive plus $250,00 pain and suffering) plus an unlimited amount of economic damages (lost wages, etc) saves about $62 billion due to reduced cost of medicare, medicaid, FEHB, and Obamacare subsidies. This strikes me as a large estimate when you consider that it only happens if the decreased malpractice liability causes both doctors to reduce prices and insurers to pass that savings to enrollees, for a net reduction in total national health spending by 0.5 percent.


If you repeal Obamacare's expenses but not it's funding sources, you have a pot of $1.2 trillion to play with.

On the other hand if you just want to end the individual mandate, that's $416 billion in savings on Obamacare subsidies as people become uninsured. To be honest, that's more than I would estimate, as my view is that the subsidies themselves are a larger incentive to stay in the market than the mandate.

12/09/2016 12:37:00 PM


Ok, there is one paper that suggests, yes maybe a little tiny bit. From the abstract:

Results indicate that consumer voters were attracted to states with higher per pupil public school spending, lower property and income tax rates, and that certain consumer-voters may be attracted to states that offer higher levels of Medicaid benefits.

This paper studied pre-ACA data and found a small marginally significant correlation between per-capita medicaid expenditure and net migration. But the paper doesn't use a particularly convincing strategy for causal identification—correlation is not causation, as they say, and in any case the correlation with medicaid was among the weakest they found in the data.

Two other papers have studied the effects of various expansions of medicaid on migration, using much stronger methods for causal inference. Here's Schwartz and Sommers (2014) studying pre-ACA expansions in several states:

Using difference-in-differences analysis of migration in expansion and control states, we found no evidence of significant migration effects. Our preferred estimate was precise enough to rule out net migration effects of larger than 1,600 people per year in an expansion state. These results suggest that migration will not be a common way for people to obtain Medicaid coverage under the current expansion and that interstate migration is not likely to be a significant source of costs for states choosing to expand their programs.

And here's Goodman (2016) with an analysis of the ACA expansion specifically:

Using an empirical model in the spirit of a difference-in-differences, this study finds that migration from non-expansion states to expansion states did not increase in 2014 relative to migration in the reverse direction. The estimates are sufficiently precise to rule out a migration effect that would meaningfully affect the number of enrollees in expansion states, which suggests that Medicaid expansion decisions do not impose a meaningful fiscal externality on other states.

Based on my brief search of the literature these are the only papers on the effect of medicaid on migration specifically, though a much larger literature exists on the effects of welfare programs more generally, usually finding small effect sizes.

The migration literature, including the first paper above, does find that tax policies have a significant effect on migration, with higher taxes causing lower net migration to those localities, consistent with the theory that people move to avoid taxes. It's not hard to guess why taxes would have much larger effects than welfare generosity: the rich who pay taxes can afford to move, while the poor who receive welfare cannot. There are two ways of interpreting this disparity in migration effects. On the one hand, the fact that medicaid doesn't induce migration relaxes incentive compatibility constraints, allowing more efficient redistribution, while on the other hand the high effects of taxation on migration exacerbates these constraints, prohibiting states from being able to finance as much redistribution as they otherwise prefer.

11/29/2016 06:10:00 PM

In my previous post I explained why Fedcoin—the idea for replacing dollar bills with a blockchain-based cryptocurrency like Bitcoin—is completely compatible with fractional reserve banking. Much of the discussion of Fedcoin has assumed that

  1. The Fed would start administering "accounts" for private individuals (they currently only administer reserve accounts for banks)
  2. Most people would prefer to use their Fed account as primary deposit accounts instead of using private banks.

I think both premises are wrong.

I suppose, in principle, the Fed could start administering deposit accounts for private citizens, but I don't see any reason they would do so. This argument is non-unique—the Fed can already do this if they really want to, and the introduction of Fedcoin doesn't substantially alter the difficulty of doing so. Online-only banking already exists, even without using any cryptocurrencies.

Given the choice between wiping out the entire banking industry and not, my bet is that the Fed will choose not.

The notion of "accounts" or "digital wallets" or "deposits" do not appear anywhere in the blockchain protocol. Raskin and Yermack (2016) seem a bit confused on this point. They say:

The blockchain is known as a "shared ledger" or "distributed ledger," because it is available to all members of the network, any one of whom can see all previous transactions into or out of other digital wallets.

And their fear that an individual's entire "digital wallet" is discoverable through the blockchain underlies John Cochrane's post lamenting the privacy implications of Fedcoin. Although the blockchain does pose some privacy challenges, Raskin, Yermack, and Chochrane overstate the problem. In fact, digital wallets are not part of the blockchain and cannot in general be reconstructed from it.

The exact technical details are beyond the scope of this post, but I think Raskin and Yermack have confused the concept of blockchain "address" with "digital wallet." Addresses are not like bank accounts—an address is really just a fancy word for a serial number, analogous to the serial number printed on each dollar bill in your wallet. Each serial number is cyrptographically related to a different number called a private key, which what you use to authorize the transfer of the bitcoins associated with that serial number to a different serial number. This transfer from one serial number to another serial number is the only information recorded in the block chain. A digital wallet is third-party software that can generate serial numbers and manage their corresponding private keys automatically on your behalf. Importantly, there is no way, using only information in the blockchain, to tell which serial numbers are being managed by which digital wallet—these are totally separate systems.

That's not to say there's no privacy concern. In a Fedcoin world, your employer would pay you by transferring Fedcoin to a serial number you control, and which you communicated to them. Thus, your employer knows that that serial number belongs to you, and they can see that serial number in the blockchain. If you then spend those Fedcoins at a porn shop, for example, and your employer also happens to know what serial number the porn shop is using to accept payments, then your employer would be able to see in the block chain that you spend your paycheck at a porn shop. But at the same time, you can anonymize your transactions pretty easily. For one, after you receive your paycheck from your employer you can transfer those Fedcoins again to serial numbers your employer doesn't know about. The employer would only see that the money was transferred—you might still have them, or you might have deposited them at a bank, or you might have spent them on rent and groceries, no one would have any way of knowing. But also, stores would most likely use different serial numbers for each transaction, so in practice the above scenario—where the employer knows which serial number the porn shop will use—wouldn't ever happen. (As an aside, note that depositing your Fedcoin in a bank enhances privacy!)

Fedcoins are not inherently anonymous, but completely anonymizable. Expect this to be a major topic from now on.

As mentioned above, to control you Fedcoin you need to retain secret private cryptographic keys, and normally you'd manage those keys with software called a digital wallet. Your digital wallet would then be secured by a password, which is up to you to remember. But humans are notoriously bad at passwords. We pick passwords that are easy to guess, but then forget them anyway. We reuse passwords accross multiple services. Many of those services fail to properly secure those passwords, leading to data breaches. We never change those passwords, unless someone makes us. And that's just password-related risk you incur by having Fedcoin in your digital wallet. You still have all the same physical risks as you have with paper-currency. If your digital wallet is stored on a harddrive that fails, you can lose your fed coin that way. Cloud storage can fail too, no matter how many times a day it gets backed up—all it takes is one poor architecture choice and one typo by a programmer (I've seen it happen). Your Fedcoins are not fire proof. Not necessarily water proof. Not even magnet proof.

With a traditional bank, our inability to secure our online lives doesn't cause too much mayhem. Banks have experts to secure their computers, and employ sophisticated fraud detection software that can freeze accounts before much money has been spent. They insure us from fraud and offer us ways to recover our accounts even when we forget our password. They accept your actual in-the-flesh person as proof enough that you are who you say you are. But Fedcoin does not come with any of these protections. Unless you put your Fedcoins in a bank, you are on your own to secure your private keys and if someone should find a way to hack them and spend your Fedcoin, there's no way to ever reverse that transaction or to find the crooks who stole them. If you forget your password for your digital wallet, there's absolutely no way to ever recover those Fedcoins. Indeed, in the case of theft or forgotten password, there wouldn't even necessarily be a way to prove that those were your Fedcoins that were stolen or lost—the lost private keys themselves are the only proof with any credibility in the blockchain world.

All of this is to say that while large sums of Fedcoin are cheaper to store than cash, they are by no means any safer. For this reason, most consumers will continue to prefer to keep their Fedcoins secured by third party institutions, which force us to use proper security in exchange for insurance against liability.

In principle, these institutions need not be depository banks—consumers could instead pay a maintenance fee. However, consumers tend to be significantly averse to fees, and they will almost certainly opt instead for institutions that make money by lending those Fedcoins out, like banks, instead of paying maintenance fees on their balances. In a world where Fedcoin pays interest, banks would need to compete with interest-bearing money and so we'd probably see banks start paying out higher interest rates on deposits. If their returns on lending out a fraction of those deposits are high enough, banks might even match the amount of interest the Fed pays on Fedcoins. However, my guess is that in practice consumers would be willing to accept a lower rate on their deposits—a premium for the insurance against various kinds of risk Fedcoin incurs.

8/15/2016 05:27:00 PM

Commentators often argue that replacing cash with "Fedcoin"—a Fed-administered cryptocurrency like bitcoin—would end fractional reserve banking. JP Koning, who wrote the cannonical piece about the idea of Fedcoin, writes

Fedcoin has the potential to tear down the private banking system...As the public shifted out of private bank deposits and into Fedcoin, banks would have to sell off their loan portfolios, the entire banking industry shrinking into irrelevance. One way to prevent this from happening would be for the Fed to make an explicit announcement that any bank could be free to create its own competing copy of Fedcoin, say WellsFargoCoin.

Koning argues that fractional reserve banking is doomed unless commercial banks are able to launch separate side-chains that build off of the official blockchain maintained by the Fed. The idea that Fedcoin would wipe out fractional reserve banking also underlies much of John Cochrane's post where he quotes Raskin and Yermack (2014) who also take this as a given:

Depositors would no longer have to rely on commercial banks to hold their checking accounts, and the government could get out of the risky deposit-insurance business. Commercial banks that wished to keep making loans would raise long-term capital in the debt and equity markets, ending the mismatch between demand deposits and long-term loans that can cause liquidity problems.

This aspect probably appeals to Cochrane because he wants to abolish fractional reserve banking as a matter of policy, to end the possibility of bank run equilibria.

But Koning, Raskin, Yermack, and Cochrane have overthought this. Taking a step back, it's clear that blockchains pose no threat to fractional reserve banking (for better or worse).

Here's the problem I think Koning had in mind when he proposed side-chains: Suppose you own some Fedcoin and you want to deposit them into a bank. One way you could "deposit" Fedcoin is by sharing your private keys for your Fedcoin addresses with the bank. This way, the bank would be able to use your keys to send some of your Fedcoin to borrowers, while you would also be able to use your private keys to use some of your Fedcoin to buy stuff. But there's an obvious problem: you can't spend Fedcoin from the same addresses the bank is lending out to others, because only one of those two transactions will clear in the blockchain, regardless of whether, in principle, the bank has enough Fedcoin deposits to cover these liabilities.

To avoid this, Koning proposes that instead of sharing your private keys with the bank, you'd sign over the Fedcoins to the bank and receive WellsFargoCoin in exchange, which stores would accept as payment because they trust Wells Fargo's guarantee of redeemability. This way, Wells Fargo is free to lend out your Fedcoin deposit. That's one way to do it.

But there's actually a much simpler option that doesn't require side-chains. When you make a deposit you'd sign over your Fedcoin to the bank, as in Koning's version, but you don't get WellsFargoCoin. Instead, just as you do with cash, you'd withdraw whatever amount of Fedcoin you want as spending money. You are then free to spend the Fedcoin you withdrew, while the bank is free to lend out the rest of your deposits.

In practice you wouldn't even be aware of the need to "withdraw" Fedcoins before spending them. Your digital Fedcoin wallet could just do this automatically. You pay for a cup of coffee by pressing a button on your phone, the digital wallet sends an order to your bank to withdraw the funds, then sends the funds to the coffee shop's Fedcoin address. No side-chains necessary.

In this post, I explained why fractional-reserve banking is simple to implement in a blockchain even without sidechains. In a future post, I'll explain why it's likely that most consumers will prefer to keep their Fedcoin in banks.

8/10/2016 09:59:00 AM