Kurt Johnson

Walking around Old San Juan, Puerto Rico last December we came across several groups of police getting their shoes shined by street vendors; this was the group that photographed best.

Canon Digital Rebel, 17-40/4L @ 17mm/8

On the recommendation of The Economist, I’m reading Global Crises, Global Solutions (Bjorn Lomborg, 2004).

The compilation identifies education as a critical issue for worldwide improvement. The readings generally show very weak correlations between increased spending on school inputs (infrastructure, teachers) and positive student outcomes, leading me down this tangent of thought: if an individual school really wants to improve its rankings/status/productivity, the efficient path is to view competition for students as a zero sum game and make investments in attracting better students instead of improving the education process. The competition for better outcomes by an individual institution is largely won or lost the moment a student commits to going to school.

p.203: “Much of school-specific variation depends on the fact that student household background characteristics vary across schools….This means that the maximum amount of student performance [on Brazil’s PISA exam] that could be explained by everything about the schools is 20% of the observed variation in student scores This poses large problems for the researcher in trying to disentangle the causes of higher student performance as modest amounts of sorting by parents and students into schools can lead to large bias in estimates of the relationship between school factors such as class size and learning outcomes.” [emphasis is author’s]

In other words, pouring funds into classrooms, salaries, materials, etc. may result in improved outcomes, but these outcomes account for less than 20% of total outcome variation. The schools is much better off NOT investing in the educational process but instead stepping up marketing, scholarships, and student amenities to attract better students (which accounts for 80% of the outcome).

Taking this to the next logical step, if we are in a reasonably free market for students to choose their school from a slate of options (such as US universities) we should see intense university recruiting competition for high quality students; anecdotally this takes the form of universities’ drive to build student recreational facilities (climbing walls, elaborate gymnasiums, movie theatres on campus). These facilities have no positive impact on education outcomes, but they are very potent in attracting more students and giving the university power to select the ‘right’ students and thus pull the 80% lever. Personally I think Harvard kind of sets the standard for courting the ‘right’ students based on their ex-academic potential (read: connections -think G.W. Bush, Harvard MBA ‘75).

The really interesting point to me is: since the 80% lever is so powerful, a school that desires to build reputation should basically drop everything and focus all investments on student recruiting. In other words, tuition should drop to zero or negative in order to attract more desirable students (i.e., grants and stipends beyond the cost of tuition). So why do the vast majority of private schools still have large positive tuitions? Why do we see declining numbers of merit scholarships? Shouldn’t ambitious universities be cutting prices like crazy to yield more desirable students? Why is merit-driven competition for students so weak? We could hypothesize several things:

• that the top schools’ history of price-fixing/cartel behavior is continuing despite court losses
• that desirable students are highly insensitive to tuition prices. This doesn’t make much sense to me either.
• that competition between top schools is already at an equilibrium point; schools have already identified the price/amenity point that maximizes their take of desirables students (i.e., the cost in offering more benefits is not justified by the gain in outcomes/rankings). Given the size of the 80% lever combined with the size of endowments at some schools I think this is unlikely
• That schools choose not to compete for students this way or compete only in a limited way because they recognize that competing for students is a ‘Red Queen’ game (or ‘arms race’) in which everyone can make large investments to compete and not realize a net gain (because outputs are zero sum). Economists would probably not believe that a market with dozens or hundreds of participants could act in such a concerted way without organized collusion (see bullet #1). The temptation for one school to break ranks and start the competition is too great (this notion is a great lead in to Edgeworth cycles if anyone wants to go learn more about game theory and price competition in a market with many substitutes)
• That there are possibly political considerations that make administrators shy away from pure-merit driven competition (e.g., disparate racial impact or throttling ‘diversity’ in terms of race/income/religious/etc.)

Obviously all of this only makes sense from the viewpoint of an individual institution in competition with other institutions. For a national-level education system a zero-sum perspective is useless.

Footnote: The article uses standardized test scores as a proxy for educational outcomes and so do I – I don’t know of any other metric which can be compared across thousands or millions of students in a meaningful way. Graduate income (controlled for incoming student SES) is a nice thought but full of problems.

Corollary: One of the interesting points in the book discusses the insignificance of investments in education as predictors for output. As the 20%/80% split above implies, it’s statistically difficult to show a meaningful bump in performance by spending more money, whether on new schools, infrastructure, or smaller class sizes. Anecdotally I think this is again supported by US universities; there doesn’t seem to be much hoopla about class sizes in undergraduate classes that commonly enroll 50-500 students (probably because professors/GAs aren’t unionized and thus don’t lobby effectively for more work the way that public schoolteachers unions do).

Apparently some folks are having problems with comments, and I just noticed that the links to archives don’t work (which is OK, sort of, because you can just scroll down to see old posts).

In an unposted comment on this post, Lucas says:

I liked your idea of the localized MBS tranches but I was
thinking, wouldn’t a futures contract based on something like the
median home price for an area make even more sense? I think real
estate prices tend to get out of whack because of the relative
illiquidity of the market, and the easy access people have to leverage
for the investment. With a futures market, retail investors would be
able to put money into the real estate market without the costs
associated with loans and real estate brokerage. Construction
companies would be able to hedge against risk when building new
subdivisions or other construction projects, and prospective home
buyers would have an investment vehicle that would allow them to
“track” the gains being made in their local housing market and thus
avoid being priced out of the market as they build savings or if they
currently live somewhere else (say while you’re in Philly, you want to
track the Dallas market with money earmarked for a down payment).

I think defining the underlying asset would be the most difficult part, since real estate is very localized (downtown Chicago has a different appreciation curve than the South Side or the Western Suburbs) but contracts for the 20 largest metro areas might have some interest. I t could even be pitched as a substitute for PMI for borrowers with low downpayments (instead of paying PMI have them sell
contracts for their real estate market)……It would probably require too much
investment saavy for a regular consumer but regional banks and
mortgage companies could possibly provide enough liquidity to sustain
a futures market.”

Regarding the futures idea, Robert Shiller has actually talked about getting housing futures going for exactly the reason Lucas suggests. Without having done any reading on how Shiller’s work or their success, I think traditional futures would be very tough to pull off. Futures work best with fungible commodities – goods where a common ‘unit’ can be produced by many suppliers and is acceptable to many consumers. Grain fits this definition, as does natural gas, gold, raw cotton, etc. Another key to futures markets is that since participants are buying and selling a contract on a physical good the futures prices ‘converge’ to the value of the physical asset even though most participants close out their futures position before expiration and delivery occurs. I don’t think Housing futures fit either of these requirements; housing values are highly location-specific, it’s hard to get a ‘fungible’ basket of physical houses that represents the same intrinsic asset from one month to the next (remember the houses in the basket need to actually available for title transfer if the futures contract is taken to delivery). However I think the basic idea (and I suspect this is how Shiller’s plan works) could be tackled by creating an index of housing transactions to support derivatives such as swaps. There is still the issue of relatively low liquidity and how to qualify transactions for the index (4 BR houses 1800-2200 SF? What about quality of schools, age of house, yard size, etc.?). Getting enough transactions into the index to make it statistically usable (hard to manipulate, accurate, etc.) might require making the breadth of geograpy, size, or amenity too wide to be useful. I agree with Lucas that participating in a futures/derivative market is probably too sophisticated for 95% of homeowners and probably not a good idea even if they thought it was. I don’t know who would be long the futures/index either; I can see homebuilders, banks, and short-holding-period homeowners taking the short side but who really wants the long side? Lucas’ thought that future homebuyers would take the longs as they build their savings actually makes a lot of sense, but I can just picture someone screaming to their congressman because they took the long side of this deal while building their savings and then lost all of the savings when the market dipped. If you think the market is going up isn’t it better just to buy a REIT/homebuilder/home/investment property?

Regarding the PMI issue: in my mind (I’m no expert) PMI is protecting lenders from a severe market downturn where people who have very little equity just walk away from their notes as in the mid 1980s oil patch. This risk could be hedged with futures/derivatives as Lucas suggests. However I think that to replace PMI the homeowner would have to basically put the futures into an escrow account – otherwise if they owned the futures outside of their mortgage contract and the housing market tanked they could basically keep the profits from the futures, walk away from their underwater mortgage and double their money. So you’d have to force the homeowner to sign over the profits from the futures (or better, a put option) to collateralize the mortgage. If that’s what you’re doing I think it’s more efficient to just keep the PMI construct – homeowner writes a check for PMI, then the PMI company shorts futures to cover their risk. It’s going to be more efficient and less controversial to have PMI companies doing the hedging instead of the homeowner (and presumably PMI companies would let their end-user rates float based on the premiums they were paying for puts).

Reflecting on what I wrote about real estate in the last post, I still really like the idea of localized MBS tranches; I think it would better let the market charge homebuyers for the local risks in their geography and thus help self regulate bubbles and depressions (the so-called ’soft landing’ from a market bubble). Tranches could be created for various locations and home types (which already exist, I believe, or at least jumbo/non-jumbo splits). Obviously to keep sufficient liquidity/fungibility in the market some tranches would be very geographically broad (all of Wyoming might be one tranche) but large metro areas should be able to support their own tranche. I don’t think I’m being presumptive when I say there’s a consensus opinion that some urban markets are in a speculative bubble with unprecedented investor leverage and large fractions of transactions being undertaken by speculators/investors rather than resident owners. Say that the San Diego $500,000-$1,500,000 mortgage market is one of these areas and we can break it out into its own MBS tranche. MBS buyers should be cautious of owning these securities (because the high homeowner leverage and speculative market raise default risk); because buyers require a higher than normal yield to hold San Diego MBS, end-user mortgage rates in the area will rise. This will cool down the housing market in the natural cycle of monetary self-regulation. I’m afraid that what’s happening today is that San Diego buyers get a nationally-determined mortgage rate and so do buyers in Wyoming; San Diego homebuyers should be paying more because they’re in a riskier market, but instead everyone gets a blended rate – Wyoming pays too much and San Diego pays too little. In theory tranching gives the market a more efficient self-regulation mechanism than the national-level aggregation we have today; it should keep hot markets from getting too hot and make it easier to buy cheap in cold markets (and get everything back to a healthy lukewarm balance of supply and demand).

Caveat: I don’t know if the MBS market already does this; I know in the early days of the MBS market it was pretty easy to see the individual notes that made up the security; once tranching really took off I think the securities got less transparent since banks had to aggregate so many more mortages into a pool in order to create the 10+ tranches that the market was demanding. This would be easy to check out if I actually had a stake in it or a Bloomberg terminal, which I don’t.

I don’t know if the Feds have laws against charging different mortgage rates in different locations (which basically says everyone has to subsidize the highest risk mortages in the country). I also don’t know how default risk actually trickles down to the MBS cashflows; presumably the vast majority of MBS cashflows are backed by Fannie Mae/Ginnie Mae so I don’t know if a default actually has any effect except altering cashflow timing (or IO/PO splits – I’m not sure if a FHMA-secured default results in a long cashflow stream according to the original contract or a lump sum prepayment of the note). This is a pretty key assumption that I’d need to understand much better before getting too worked up about anything.

Sorry for the length of the post and lack of supporting documentation – I’m stuck on an airplane again with no internet access and nothing better to do.

Maybe the best quote I've seen on the folly of US News (and other) rankings for competitive schools:

Robert L. Woodbury, former chancellor of the University of Maine system, noted the folly of the current institutional U.S. News and World Report rankings: When Consumer Reports rates and compares cars, it measures them on the basis of categories such as performance, safety, reliability, and value. It tries to measure “outputs”—in short, what the car does. U.S. News mostly looks at “inputs” (money spent, class size, test scores of students, degrees held by faculty), rather than assessing what the college or university actually accomplishes for students over the lives of their enrollment. If Consumer Reports functioned like U.S. News, it would rank cars on the amount of steel and plastic used in their construction, the opinions of competing car dealers, the driving skills of customers, the percentage of managers and sales people with MBAs, and the sticker price on the vehicle (the higher, the better).

For a better ranking system, here's a link to an excellent article (at Wharton, no less ;-). It outlines how to build a market based ranking system and illustrates more of the flaws in the current paradigm of college rankings. Like a lot of economics it's generally intuitive but the detailed data opens some questions at the margins.

This graph from the paper showing the dip in Princeton admissions rates for students with SAT scores between the 93rd and 98th percentiles is a great example the distortion that rankings regimes can cause:

Ultimately I think understanding a consumer decision-based ranking could really help the schools shape their classes a lot more effectively than the crude rankings do today. For example, BYU has cultivated a strong preference in a subset of applicants (p. 38) that disproportionately prefer BYU to other normally more desirable schools. Understanding what segments of applicants strongly favor or disfavor their institutions should allow admissions committees to more effectively market to these segments in order to shape their class – and ultimately develop their university and brand – on the demand side (applications) instead of the supply side (admissions).I can't decide whether or not taking up this approach (vs the current rankings) would change the way that students apply to schools. Certainly Princeton's example is a little disheartening for the 96th percentile student; if applicants were armed with more detailed admissions figures the 96th percentile students might apply in lower numbers to Princeton (ceterus paribus) which would harm the school's applicant count -> selectivity % -> ranking under the current regime. Thus Princeton could be expected to fight more disclosure in this case.

Another question I haven't resolved is whether a revealed preference/market ranking would do more harm than good. It's certainly more empirically valid in terms of a preference ranking, but I think it's also clear that some schools coast on a reputation earned a long time ago; the brand of a school may lag what that school is actually producing/providing by a generation. In that sense a revealed preference ranking is a trailing indicator that isn't really useful for anything except a prestige index; for a forward looking student or a university trying to pull itself up by its bootstraps a ranking system has to measure more immediate indicators of output quality. I'll draw an analogy to my view on standardized testing for students; I don't think teaching to the test is bad per se, it just means you have to be very careful to structure the test so that teaching to it effects your ultimate teaching goals. Likewise, rankings parameters should be structured so that managing to rankings metrics serves a positive good. Class size, outgoing placement rates/salary, and ROI are good things to manage to while the current metric for 'selectivity' or acceptance percentages is destructive.

Obviously I’m very excited about the news.

I’ll offer more commentary in a few weeks after the Columbia process wraps up. And, although hopefully this is repetitive, thank you to everyone who has advised and supported me through the applications gauntlet.

Canon 20d, 17-40/4L @ 17/4, 1/15 second, ISO 400

Wharton decisions come out tomorrow morning. Columbia just invited me
to interview, which hopefully indicates a decision will come within
the next 3-4 weeks.

Last christmas my wife upgraded my DSLR to a Canon 20d. One tangible benefit was a huge jump in speed; my 300d took 2.5 shots/second for up to 4 shots in a row while the new camera focuses faster and takes 5 shots/second up to 25+ frames before slowing down. This allows for some shots I wouldn’t have gotten before, but it’s also made me lazy about taking time to set up shots properly – it’s too easy to just rip off a bracketed series of frames and plan on having one turn out. Overall I think I was better off when I was forced to slow down, and I’m making an effort to be more disciplined about using the speed of the new body.

I’m also questioning my investment in full-frame Canon EF lenses instead of a reduced frame (EF-S) line. The more I think about it the less likely I think I am to need a full-frame sensor; the 1.6 crop factor on my 20d gives adequate resolution and low enough noise for everything I need, and certainly the next generation of DSLRs will be even better. Unless I’ve got a real need for >18″ prints/8-10 megapixels what do I gain by going to a larger sensor? Ergo, why am I hauling around the extra weight/expense for the full frame lenses when I could get away with reduced frame equivalents? I’d be more inclined to actually move on this – sell my EF lenses and replace them with EF-S – if I knew that Canon was committed to another 2+ generations of reduced frame DSLRs (to ensure a long life cycle for my lens purchases) and if the EF-S lens offerings were as good in quality as the full frame EF/L lenses I’ve got now…..but I’m still thinking about dumping my 50/1.8, 17-40/4L, and 28-135IS to replace them with a 17-85IS (or maybe Sigma 18-50/2.8?) and maybe a 10-22….Just idle thoughts for now.

I’m also looking for a better solution to creating decent-looking web albums. So far I’ve tried:

• Photoshop’s native web album automation – Disappointing flexibility to customize layout, upper limit on image size is too small for me (~480 pixels height or width?)
• AlbumGV 1.7 – Nice clean layout, built in sharpening option for downsized images, customizable output JPG quality and decent layout flexibility. I used this for most albums on my site but am moving away from it because it won’t display EXIF info
• jAlbum – Look/feel of albums is extremely customizable but conversion (sharpening) options seemed lacking and Porta was easier to get started with.
• Porta – What I’ve used for the most recent 6-8 albums. Lots of flexibility for layout, allows automatic sharpening of downsized images, displays EXIF data in a mouseover format (nice). However sharpening seems overdone and there’s no way to adjust it. I miss the functionality from AlbumGV that lets me set a maximum height for the image and let the width float (since on most displays the limiting factor is height regardless of width)
• Gallery – Too complicated for me to run on my simple hosting solution (readyhosting.com)

So basically I want a solution that outputs HTML/JS code (not active server type solutions), allows me to customize JPG quality and sharpness parameters, allows me to set a fixed height per image and let width float, and displays EXIF info from the original file. Let me know if you’ve got any ideas.