Thursday, April 23, 2015

Maybe there is no twitter trading algorithm

Slate has the story of a wall street fund manager who's convinced someone has invented a hyper-intelligent stock trading algorithm that can read and understand tweets* and headlines and make stock trades based on them:
"He’s convinced someone’s figured out an algorithm that’s faster than anything he’s ever seen before. So fast, he fears, that it might eventually put him out of a job."
*Ok, the idea that someone had programmed a bot to read tweets became something of an urban legend recently, though Reuters showed that the specific example the stories pointed to was actually from a news headline, a few seconds before the tweet. But whether it's a news headline or a tweet is hardly the point.

I'm going to make a bold prediction: no such algorithm exists. While I generally side with Alan Turing's view--the fact that humans can do something is evidence that a computer program could do the same, given sufficient resources--such an algorithm would be enough to fill several doctoral dissertations and get you a phd from any university you choose. If there was a Nobel prize in computer science, such an algorithm would win you that.

On the other hand, it would be exceptionally easy to write an algorithm that looks like one that is trading based on news headlines and tweets.

Think about it. Insider trading is illegal, which in practice means that you can't act on any insider information until that information is officially made public--otherwise, you'd get caught and prosecuted. But if you have insider information about something like the Altera-Intel deal, then it would be exceedingly easy to write a web-scrapping program that looks for published headlines and tweets about it, and automatically make the trade once the headline is found. You'd earn about the same returns as with a pure insider trade, but if anyone asks you can plausibly claim to have learned the information--or rather, your bot learned the information--from a publicly available source.

So, there's no hyper-intelligent news-wire-reading bot. Far more likely that we're witnessing a new spin on the same old insider trading game.

Monday, April 20, 2015

Can this really save money?

Chris Christie just re-entered the 2016 presidential contest with a speech about a plan to reform medicare, medicaid, and social security. The stuff about social security was the same old crap: benefit reductions justified by the circular logic that if we cut benefits now we can stave off an eventual automatic benefit cut, the reasoning that "fairness" requires that we force today's young to pay full benefits to today's retirees for the privilege of getting only a fraction of those benefits in the future, and a string of proposals that, in any case, don't actually reduce outlays that much.

But it was something that Christie said about Medicare that caught my attention:
"For Medicare, the rules for cost sharing– deductibles, caps on coverage, and co-pays, are all over the map. There are different types of deductibles and cost-sharing depending on the type of service provided and where it is provided. We should do exactly what the bipartisan Simpson-Bowles report – a report which President Obama ignored -proposed. Simplify all this by having a uniform single annual deductible of $550, a uniform coinsurance rate of 20%, 5% cost-sharing above out of pocket expenses of $5,500 in a year, and a hard cap of $7,500 a year in cost-sharing. This makes sense to me. It is simpler. It will save money. We should do it."
I won't claim to have done the math to see how close to optimal those numbers are--when you apply optimal control to health insurance, you generally don't get a coinsurance rate that's anywhere close to linear, especially when incentives for things like preventive care are considered. But what I like about this is that it ultimately caps senior's liabilities at $7,500 per year. Debates about health insurance tend to focus almost exclusively at what goes on in the truncated left-tail of the distribution--deductibles, co-payments, coverage for preventive care, co-insurance rates on small medical bills--but in my view limiting enrollees' right-tail liabilities is far more important than anything that goes on in the left tail.

The current Medicare program doesn't limit right-tail liabilities. Medicare part B, covering out-patient services which are growing portion of total expenses, has a flat 20 percent co-insurance rate on all spending above the deductible. This scheme was designed when outpatient care wasn't very common and generally only involved small inexpensive procedures. But we now do things like surgery on an out-patient basis, and the bills can easily be in the tens of thousands or more--meaning 20 percent packs quite a punch these days. Medicare part A is much more generous than part B on the left-tail of the distribution, with low coinsurance rates on hospitalizations...as long as you don't spend too many days in the hospital. In fact, part A cost-sharing rises steeply as your inpatient expenses rise, and benefits drop off completely at the 90-day mark (well, you get an additional 60 "lifetime reserve days" that don't renew). These benefit schedules might seem optimal in the neighborhood of average medical costs, but they fail to protect seniors from the biggest risks and are probably deeply sub-optimal as a result.

Christie's plan fixes that by capping out-of-pocket costs at $7,500 per year.

But here's the thing: can this possibly save money? I'm skeptical. A cap on out-of-pocket expenses is substantially more generous than the current system's treatment of right-tail risks, and the 20 percent coinsurance rate is no higher than the current 20 percent coinsurance for part B. The deductible is actually less than half the current medicare part A deductible. Where are the savings?

While I don't really know the answer, here's a few guesses. Most of the savings comes from higher out-of-pocket costs for inpatient hospital stays. Current inpatient stays have low co-pays, and Christie's plan would replace that with a high coinsurance up to the $7,500 limit. Even short inpatient stays are quite expensive--a lot of seniors would end up paying $7,500 almost every year (for reference, average annual Social Security payment is about $14,000). Smaller savings would come from the fact that Christie's deductible is higher than the current part B deductible, as well as the repeal of part B's extra benefits for things like preventive care, which currently come at little to no cost out-of-pocket.

Even so, I'm a little skeptical this would actually save money on net. $7,500 is high compared to seniors' incomes, but awfully low compared to typical medical bills.

Friday, April 17, 2015

The irrefutability of neo-fisherism

Here's a question I've been thinking about: how will we ever know if the neo-fisherite hypothesis is correct?

The neo-fisherite hypothesis is one of the alternative macroeconomic models to emerge after the Great Recession, and it holds that the conventional Taylor Rule is exactly backwards: lowering interest rates actually causes inflation to fall and raising the interest rate causes inflation to rise. John Cochrane has a nice exploration of the idea here. The idea itself isn't new--it has been in the back of macroeconomists' minds since the early days, though I'd credit Noah Smith with giving the hypothesis it's name. The difference between neo-fisherism and orthodoxy is surprisingly subtle--they rely on the exact same equations, and differ only in one particular assumption about the direction of causality. In the neo-fisherite view, causality runs directly from interest rates to inflation so that setting interest rates is the same as setting the inflation rate, whereas in the orthodox view interest rates have only an indirect effect on inflation through monetary operations, while direct causality runs from inflation to interest rates as central banks react to changes in the economy. The difference is so subtle that many macroeconomists aren't even aware they are making these assumptions--a problem worsened by the unfortunate habit of New Keynesians to leave the money market implicit (a topic for another post, but point is there are two equations in NK models that short-circuit the money market, but the model is equivalent to a money-in-utility type model where the central bank uses Open Market Operations to set interest rates).

So the two models differ only by the assumption of the direction of causality between inflation and policy interest rates. And the policy implications of each are exact opposites. How will we know which is right? Reverse-causality is a classic empirical trap.

Here's the dilemma. Let's compare the New Keynesian and Neo-Fisherite predictions right now.
  • New Keynesian prediction:
    As the economy improves and inflation picks up, the Fed will raise interest rates
  • Neo-Fisherite prediction:
    as the Fed raises interest rates, the economy will improve and inflation will pick up
Despite having opposite recommendations for what the Fed should do, those are both the exact same prediction!

Thursday, April 16, 2015

Does welfare for workers subsidize Walmart?

This is a claim that has become a popular talking point on the left: by paying workers so little that many of them qualify for things like medicaid, TANF, food stamps, and EITC, big employers like Walmart are essentially being subsidized by the welfare system.

At one level, I think most proponents of this argument are making a values claim and not a causal one. Compared to the counter-factual where Walmart and others paid enough so that none of their employees qualified for public assistance, these companies are the beneficiaries of huge public subsidies. That's not the same as arguing that simply repealing welfare would be enough to cause these employers to pay that much. The general sense is that Walmart and others could pay all their workers enough:
At over $446 billion per year, Walmart is the third highest revenue grossing corporation in the world. Walmart earns over $15 billion per year in pure profit and pays its executives handsomely. In 2011, Walmart CEO Mike Duke – already a millionaire a dozen times over – received an $18.1 million compensation package. The Walton family controlling over 48 percent of the corporation through stock ownership does even better. Together, members of the Walton family are worth in excess of $102 billion – which makes them one of the richest families in the world.
not necessarily that they would in a world without public subsidies:
Meanwhile, Walmart routinely blocks any attempt by workers to organize, using anti-union propaganda and scare tactics, firing employees without just cause, failing to provide any form of decent healthcare coverage or a livable wage.

To make matters worse, these abusive Walmart policies have increased employee reliance on government assistance and the need for a government funded social safety net. In fact, Walmart has become the number one driver behind the growing use of food stamps in the United States with "as many as 80 percent of workers in Wal-Mart stores using food stamps."
See the difference?

To be sure, I don't buy the values argument here. Yes, society does have a moral obligation to both ensure that everyone earns more than subsistence and to share prosperity as broadly as possible. Sometimes it may be more economically efficient to do this through policies that affect wages than through government-budgeted redistribution (though I will scrutinize your model if you make this claim), but I see no inherent moral reason to prefer redistribution through the market wage mechanism. Low wages supplemented by welfare is perfectly moral policy for low-skill workers, and the argument that Walmart is shamefully exploiting the welfare system only plays into the right-wing narrative that welfare is shameful and low-skill workers are icky. The left-wing attempt to vilify Walmart for hiring low-productivity workers only further vilifies low-productivity workers.

But then, not all proponents of the welfare-for-walmart theory are sticking to the values claim. Some go further to make the causal claim that reducing welfare benefits would cause employers to increase wages:Unlike the values claim before, this causal claim is within economists' domain.

So, if that's the hypothesis, what's the model? I think this is the model most proponents have in mind: In economic terms, welfare increases labor supply by literally keeping more workers alive, and because the labor demand curve is downward sloping this implies lower market wages. Graphically:
Welfare-for-walmart theorists believe that by keeping more workers from starving to death, welfare programs increase labor supply and thereby decrease the wages firms have to pay.
Health economics research vaguely supports the broad proposition: social factors like income have big effects on health and we have every reason to believe that welfare assistance reduces the overall mortality rate so that there are, in fact, more workers with a welfare system than without. But, we're talking about effects on different orders of magnitude, and the welfare-for-Walmart claim--that all the workers would starve to death without public assistance--is not consistent with the research finding that lower incomes decreases life expectancy somewhat. I have no doubt that some Walmart workers go hungry sometimes, but the idea that most of them would starve to death in the absence of public assistance is pretty hysterical. The american poor still rank among the wealthiest compared to the long history of human subsistence.

So, that's the theory anyway. I'm not sure it's even internally consistent--more workers alive would increase labor supply, that's true, but it also increases the demand for goods and services. In the standard DSGE model, for example, an increase in the labor supply will eventually (pretty rapidly, actually) cause the capital stock to increase by just enough to maintain the original wage rate. So without some behind-the-scenes theory for why public welfare decreases the capital/labor ratio (distortionary taxation?) this theory doesn't add up.1

There are variants of the labor supply story that don't assume mass starvation:Dube's argument here is that various welfare schemes each impose their own distortions on the labor supply decision--the EITC, for example, is essentially a tax on non-employment since you can't qualify without a job, while SNAP (food stamp) benefits which decrease as your market income rises is essentially a tax on working more. Certain kinds of tax distortions can drive a wedge in the capital/labor ratio and reduce long-run wages, but it's not guaranteed, so this theory is still missing something. Moreover, while exact effects depend on the design of each program, I don't think it's controversial to say that the combined net effect of all these welfare programs is to decrease aggregate labor supply, which would tend to increase not suppress market wages.


1. To see why I make this claim, consider the DSGE model I posted previously. (I'm posting math as images with white text--if your using an RSS feed with a white background, click over here) The consumption Euler equation without distortionary taxation is , while from the firm's profit-maximization problem . That means the steady state labor/capital ratio, , is fixed by technology, time preferences, and capital depreciation. Also from the firm's problem, , with fixed L/K implies increasing labor supply does not decrease wages, except in the short-run transition phase. No doubt there are a variety of deviations from the standard model that would allow long-run wages to change, but, as always, if you don't present your model it didn't happen. We can't debate claims that aren't stated.

Tuesday, April 7, 2015

How many people would lose health insurance if King wins?

You've probably heard of King v Burwell, the lawsuit before the Supreme Court that could take ACA subsidies away from 34 states if the plaintiff, King, wins. You're probably also aware that in such an event there is no viable plan to pass a Congressional fix and, realistically, little chance that states will be able to establish their own exchange in time to keep the subsidies.

It would be unusual if the Court stayed the ruling--though Justice Alito hinted at it--so if King wins that probably means 9,346,000 people lose subsidies for their health insurance immediately. How many will lose insurance completely?

That is a question with several dimensions. Without the subsidies, many people won't want to pay for coverage anymore at the new higher price. Then there is an additional effect caused by adverse selection: those who do still want insurance at the higher price will be the higher-cost, sicker patients, meaning that the risk pool will become higher-cost, further driving up premiums for everyone above and beyond the current pre-tax prices. RAND says premiums would rise 47 percent above the current unsubsidized rates due to adverse selection. It doesn't stop there. Those who do continue to buy insurance will downgrade to the cheapest plans offering the least amount of coverage.

Rand estimates that 9.6 million people would lose insurance entirely in this scenario, about 70 percent of the non-group market for the affected states. The Urban Institute's estimates don't look much better: 8.2 million lose insurance. And I'd add that much of the loss is likely to be permanent, as those who get burned by a sudden, massive, and totally unexpected (more than half of the public is totally unaware of King v Burwell, and even fewer know it affects their state) rate hikes will never return to the exchanges again.

What got me thinking about this was this paper from 2009 that estimated the elasticities of demand for health insurance among the self-employed. The elasticities aren't necessarily generalizeable to the ACA--theory tells us that insurance equilibria aren't very robust and depend heavily on market structure. But out of curiosity I did the calculation. They estimated an extensive (the decision whether or not to buy insurance) elasticity of -0.3 and an intensive (decision of how much to buy) elasticity of -0.7, which is pretty moderate as far as published estimates go. Just counting the loss of the subsidies, that works out to 7,205,766 people losing insurance due directly to the loss of the subsidy. If we borrow RAND's estimate of the price effect from adverse selection, that's more than a million additional people loosing insurance, for a total of 8,813,166 loosing insurance (see update). So, RAND's and Urban Institute's estimates look pretty reasonable to me.

And what about the intensive margin? The ACA metal tiers didn't exist in the study sample, of course, so we have no idea what will happen. But if we do apply the -0.7 estimate literally, that means a King victory will totally wipe out all but the very cheapest plan in the Federal Exchanges.


Update: The 8,813,166 figure has been revised up because I originally forgot to include exchange plans that experience a price hike from adverse selection but were not receiving subsidies. I'm still probably missing people because the ACA also requires insurers to pool risks across all conforming plans, including both those sold on and off the exchanges. Thus most of the off-exchange non-group market would experience a similar 47 percent rate hike.