Econometric testing

I've been reading through a metrics text and thinking about traditional significance testing. One of the problems is that you can data mine your way to significance regardless of whether or not your actual theory explains what's happening. Other tests compare two different estimators to determine if one does a better job of modeling the data than another.

This got me thinking. How about starting with a theory, creating a model based on that theory and use your data to estimate the parameters of that function. Next, set that aside and throw raw statistics at your data set to find the best estimator (or a few best estimators, based on multiple criteria) to model that data, feel free to fold, spindle and mutilate as much as your heart pleases, use parametric and nonparametric estimators, whatever functional forms you can think of, whatever, just get a good R^2 and high p-values.

For example, you are trying to estimate a firm's production function with a large chunk of data on inputs, factor prices, costs, output, etc. Theory might lead you to believe the firm  has a Cobb-Douglas production function of the form y = A*K^(p)*L^(1-p), and you use your data to estimate a, K, L, and P. Next, you pull out all the stops and try as many models as possible to use the data available to estimate y most accurately, regardless of the functional form or transformations necessary to get "good" results. This second model becomes your null hypothesis against which you test your Cobb-Douglas estimate on a new data set.

Finally, the real test is to take your theory-based model and your statistics-based models, and apply them to a new data set (preferable with at least some observations out of sample compared to what you created the original models with.) Accept your theory based model if it does a better job predicting the dependent variable than the purely statistical models, reject if it doesn't, and think carefully about it if you get a mixed bag of results. This would seem to cut down on type I errors, but would it give you more type II errors? I'm thinking it would act something like a traditional significance test, except we replace the null hypothesis of 0 with a null hypothesis of a data mined model, and I would think that the result would be somewhat more informative.

MN Government spending as a percent of GDP

Data from www.usgovernmentspending.com.

Pulled data from MN's budget office and the BEA for GDP data. because the earlier data don't go back very far. The spending numbers don't match up with the usgovernmentspending.com site, but the MN budget info isn't clear, there are "general fund" spending and "all spending" sections, both of which are smaller than what the other site showed.

Evidence of Rational Criminals

As gas prices rise, it seems like every news source I see is talking about increases in drive offs from gas stations and stolen gas. This sounds like reasonably strong evidence of rational criminals: As the price of gas rises, people are substituting from "gasoline" to "stolen gasoline." Given the number of gallons sold, number of customers, number of gallons stolen, and number of thieves at each price per gallon, you could probably estimate a cross price elasticity for stolen gas relative to purchased gas. Prepay and pay-at-the-pump stations might be confounding factors. Hmmm, it might also be interesting to see see if the relative price of gas compared to other stations is more important than the absolute price of gas in determining how much is stolen. And does everyone who steals gas fill up completely (or steal more gas on average than paying customers buy)?

From that, you could also estimate the marginal criminal's estimate of the expected cost of stealing gas. Add in the fine for stealing gas and the likelihood of getting caught, you should be able to estimate how risk loving the marginal criminal is.

I'd love to get my hands on some numbers.

Finally someone talking about Excess Reserves

We've had, what, 3 explicit rounds of quantitative easing now, as well as other rounds of less advertised expansions of the monetary base since 2007 yet we've seen no real inflation. I hear a lot of people talking about the increased M1, but very few people talk about the cause of the missing inflation: Increased excess reserved held by banks most likely because the Fed is now paying interest on reserves. Carpe Diem has a great graph:

Finally there is some discussion of this factor, according to Carpe Diem the NY Fed has a paper out titled "Why Are Banks Holding So Many Excess Reserves?"

From their conclusion:

Paying interest
on reserves allows a central bank to maintain its influence over market interest rates independent
of the quantity of reserves created by its liquidity facilities. The central bank can then let the size
of these facilities be determined by conditions in the financial sector, while setting its target for
the short-term interest rate based on macroeconomic conditions. This ability to separate
monetary policy from the quantity of bank reserves is particularly important during the recovery
from a financial crisis. If inflationary pressures begin to appear while the liquidity facilities are
still in use, the central bank can use its interest-on-reserves policy to raise interest rates without
necessarily removing all of the reserves created by the facilities.

This sounds to me like the rounds of quantitative easing, then, were not so much aimed at stimulating economic output, but were instead intended to shore up banks. You increase liquidity, and you increase bank revenue streams without significantly impacting anything outside of the banking sector... Sounds pretty similar to TARP but with less obvious numbers.

I haven't had a chance to read the paper yet, but I'm curious to see how far off the mark I am. And if this doesn't make for a subsidy to banks, what was the point of increasing the currency base without increasing the interest rate or price level?

Intrafirm Coasian bargaining

There are some situations even within firms where one department will pay the costs of an activity and another will reap the benefits. I imagine this can lead to as inefficient outcomes as out in the world when people face externalities.

Take IT support, for example. Generally this is its own department, it is staffed and funded to provide support (and often the IT related capital equipment) for the rest of the company. In large part, the budget of this department is based upon the costs required to support the needs of the rest of the firm. This can be fairly straight forward: A helpdesk that fields N calls a day needs X operators to support the volume, etc.

But not all costs are directly passed through and who determines the level of support? Take vendor support contracts for example. The IT department often pays for that contract, but they have multiple options on the service level: Should they get 24x7 support with a 4 hour response time, or 8x5 support with next business day response time? The value of the different options are based on the section of the business being supported, but the people buying the contract don't have direct access to that information. How costly is it for a site to be down overnight? That depends not only on how much business that site does, but the hours of operation. Is 24x7 support really superior to 8x5 support when the site is only open 8am to 5pm Monday through Friday?

Questions like this might best be addressed in a Coasian light: In the support contract example, rather than have the IT department handle everything, let IT use their experience negotiating the contract but have the supported business unit pay for it. That pushes the cost of the support down to the group that actually gains value from its exercise.

And when the IT support staff has to pass on the bad news that users will be down for the next couple days because the support contract sucks, at least they can blame somebody else =)

A new kind of bookstore

Maybe more of a value added service to a coffee shop. I'm thinking of a bookstore where most of the books are explicitly for browsing and reading, like in a library. Provide as much space for relaxing with a good read as shelf space, offer coffee and snacks.

Then sell ebooks. (Sure, keep a limited selection of paper books for sale directly, but focus your space on creating a good reading environment, not cramming in as many titles as possible.) Provide wireless internet connection and work with ebook providers to give you a cut on every sale from your IP address range. Sell ebooks from the register through a pay-now-download-later setup, and conversely, allow customers on your LAN to place their drink orders wirelessly.

It might not work well primarily as a bookstore, I really don't see ebooks carrying the same kind of margins as print books. But as a coffee shop, it might give you a competetive edge. And a chance to offer people a way to "try before you buy" comfortably and guiltlessly. A combination current bookstores never seem to manage.

Thought I found another great Japanese beer

Japan, East Asia in general, in my experience doesn't brew very good beer. Most of the beers I see from Asian countries are in the Budweiser territory, you drink them because you're thirsty, or to get drunk. They're mild and inoffensive, but they don't have a whole lot of flavor to recommend them, either.

Except for Hitachino Nest. This Japanese craft brewery does it right, they're easily up there with the best American and Belgian beers I've had. I would say it's my favorite brewery, and I certainly hope they expand production because at $6.50 to $9.00 a 12 oz. bottle in the liquor store, it's a rare treat.

This weekend I thought I found a comparable Japanese beer, Morimoto Soba Ale. Turns out that it's produced by Rogue (so I don't read labels closely while shopping, I like to call it minimizing search costs) so my dreams of a second excellent Japanese brewery were put on hold, but the beer is still excellent. Soba is buckwheat, so I expected something like a wheat beer. This was closer to an amber ale, none of the almost sweetness you get with wheat beer, but lighter in the mid palette than your average amber ale. Excellent from the first taste.

Add in Stone Brewery's Levitation Ale and it's been a good beer weekend. Granite City's Brother Benedict Bock is, I'm sure, a fine example of it's style but I mainly had it to confirm that bocks just aren't my thing. It's been a few years, tastes evolve, just wanted to make sure my impression was still relevant.

The news gets economics right

Just watched a segment asking "Who sets the gas price?" on the evening news. They interviewed a couple gas station owners and came to the conclusion that competition sets the prices at the pump. Best response to asking who sets the price: "The guy down the street."

Individually owned stores get wholesale prices updated daily, but their main pricing strategy is to keep tabs on their competitors and move with them. Corporate owned chains like Super America tend to move first and apparently under the direction of the parent company. Of course, I'm not sure if this is better evidence of Hayekian diffused information of Keynsian animal spirits, but I love learning about how things work.

Antiques Roadshow

It's addictive, but a couple questions always get me thinking.

First, you always hear someone crow about what a good investment something was when they hear their $40 painting from 1825 is now worth $5,000. One of these days I'll have to write down some of the numbers and see what the actual rate of return they get.

Secondly, and more interestingly: The experts almost always give a very wide price range. Appraisals like $8,000 to $12,000 are almost standard, and ranges like $2,500 to $5,000 are common as well. Does the market fluctuate that much, or is there that much uncertainty in the value that even experts can't give precise estimates?

How does securitization work?

Lately, I've been seeing comments on the financial crisis that mention securitization and pooling loans as a means to reduce risk, but treating it as a black box and not explaining how they were meant to do so. The documentary Inside Job is pretty egregious at this.

So, here's a very basic primer on how securitization works. Let's assume a bank makes 1,000 loans for $900 each on January 1st. There is a 90% repayment rate, and the amount due will be $1,000 on January 31st. There will be no profit in any stage of the process.

*Note that even though the bank is charging 11% annual interest, they are making 0 profit. This is one of the important functions of interest: To account for risk.*

So, the bank has $900,000 tied up in loans with a face value of $1,000,000. Enter the investment banker, on February 1 the banker offers to buy all 1,000 loans for $900,000. The bank likes this because it frees up that capital to be re-lent immediately, and it reduces the risk that more than 10% of the loans will default.

Now comes securitization. The investment bank puts those loans in a vault and creates a group of assets with a total value of $1,000,000. These are calles "tranches" rhymes with launches). The IBank creates two tranches worth $400,000 each (call them A abd B) and one tranche worth $200,000 (call it C). When the loan repayments come in at the end of the year, the money will first be payed to tranche A until it is completely paid off, then B will be paid, and anything left over will be paid to C.

The IBank then creates bonds that reflect these tranches and walk across the street to a rating agency. This is important because many of the institutions that buy large bonds are required by law to only buy bonds that have received a AAA rating from an accredited ratings agency. )In other words, the law gives institutional investors an excuse to shop their due diligence out to someone else.) The ratings agency will look at the structure and say "Over the last 5 years repayment on these types of loans has been 90% +/- 2%. So Bond A is definitely investment grade, AAA. Bond B is also AAA because the situation where it wouldn't get repaid has never happened in the past nor can we see it happening in the future. Bond C is only worth $100,000, and it's a B rating at that."

So now the investment bank has 3 bonds to sell. It sells bond A to a pension fund for $400,000, bond B to an insurance company for $400,000, and bond C to a hedge fund that is willing to gamble for $100,000. If 901 loans are paid off, the hedge fund will win, if 899 get repaid, it loses.

December 31st rolls around and the checks come in the mail. The IBank opens the first 400 envelops and sends the checks to Bond A's owners, it opens the second 400 envelops and sends those checks to Bond B's owner. It opens the rest of the envelopes and sends those checks to bond C's owners. Note that there is no relationship between any individual loan and which bond it pays off, the investment bank still has all the loans sitting in its vault.

So, how does securitization like this reduce risk? Overall risk isn't changed, only 90% of the loans will be paid off. But it allows the risk to be shunted down to investors who are more willing to take it, and that is used as a cushion to insulate more risk averse customers. That is, while only 90% of the loans will be repaid, securitization allows you to create a bond for 50% of the total value, and _that bond_ is very low risk: Repayments would have to be 5 times worse than expected in order for that bond to even think about losing value.

We went wrong by misjudging the risks. The models underestimated the effect of falling housing prices on mortgage defaults, and there was an assumption that the housing market wasn't national, so housing prices wouldn't fall everywhere at the same time. When defaults skyrocketed beyond all expectations, that meant that the lower, risky tranches were wiped out but higher level, less risky tranches also took hits or were even wiped out as well. This hurt institutional investors who weren't prepared for those kinds of losses. If ratings agencies or investors had been more pessimistic when estimating risk this recession wouldn't have been nearly as painful. We would have lost just as much money, but the people who lost it would have been better prepared to deal with it, it wouldn't have been as shocking and scary.

This is a very simple explanation, factors like profits, prepayments, heterogeneous loans and borrowers, and a million other factors combine to make securitization a complex mathematical jumble. But the main thing that went wrong (this time) is that the risk of defaults was underestimated.

WikEcon

I've come up with a scathingly brilliant idea, or at least interesting experiment: Create a textbook out of the Wikipedia entries for the different principles explained in an economics text. I'm thinking the format might be a paragraph or two explaining how the article fits in with the other material around it, then a link to the article, a problem set based on the information in it, and then some links to supplementary material like the original paper explaining this concept and links to Youtube videos of people teaching it or examples of it.

The interesting part, in my mind, is just how useful the wiki articles can be for teaching. A lot of the articles I've seen go into too much technical detail to be of much value in a principles text, but the nature of wiki's editing leaves it too open to inaccuracies to be comfortable as an advanced text. Maybe intermediate micro? Or it might just not work as a standalone text.

And how should I organize it and choose the topics? Actually, how does anyone organize and choose topics for a textbook? Is there a list somewhere? Right now I'm thinking of scavenging all the Half Price Books stores in the Cities for micro textbooks, creating a list from their tables of contents, and then basing the text on what they have in common plus what's interesting. Not sure that isn't plagiarism, though.

In any case, if I try this and get anywhere with it, it will certainly be a learning experience.