A highly irregular monoblog
I recently received an email inviting me to check out OnlinePhd.org, a site offering a whole load of resources for PhD students (current, prospective and past) – everything from whether a PhD is worth it in the first place, to getting tenure. Definitely worth a look for any PhD students reading this.
NB you’ll note that this is the first post here in a seriously long time. I’m considering adding the occasional post now and then, so watch this space…
Posted in /etc.
– June 16, 2012
Time for another cautious foray into the murky depths of behavioural economics! Behavioural econ covers a wide range of departures from the rational agent model and an earlier post looked at the issue of reference-dependent preferences.
Another area of research in the field is “social preferences”. Traditional economic theories assume that each consumer only cares about their own utility (happiness) and are entirely selfish; if they help someone else then it’s only to further their own gains in the long-term. Nevertheless we observe examples of altruistic behaviour which are very difficult to explain within this model: People donate money to people in far-away countries where there is no realistic chance of direct gain through reciprocation or other means.
Models of social preferences allow agents’ utility to be affected by the distribution of wealth across the entire economy. There are many possible ways in which individuals could take others’ wealth into account and a few possibilities include:
You may have noticed an interesting omission in these theories: If reference-dependence is so important, why are consumers only interested in the absolute wealth of others? Why not care about whether they’ve recently lost or gained wealth? Such a synthesis of reference-dependent and social preference theories seems a logical area for research, which also some practical applications: For example, do people prefer to give to a charity helping those in absolute poverty, or those who have suffered a sudden disaster resulting in a loss of wealth? Unfortunately there are many other variables involved here (e.g. media coverage / identification with victims) so in order to see whether reference-dependence is important in social preferences we need a more controlled setting.
Behavioural economists are fond of carrying out “lab experiments” – get a group of people, set some simple institutional rules, and see how they react. It’s true that the artificial setting may distort responses, however it is still often possible to get insights into people’s behaviour. For example there’s the “Dictator Game”: One person gets given £10 and gets to decide, on their own, how to split it with someone they don’t know and will never see again. That people don’t tend to take the whole lot is evidence against the standard selfish model.
Unfortunately it’s very hard to test for social reference-dependence in a lab setting as it’s hard to know what people’s reference points are. Nevertheless traditional theories are potentially falsifiable. Consider two players in a dictator game, called Alice and Bob. They’re both endowed with £10 and Alice has to choose a (13,10) or (15,5) payoff (where the first figure in the brackets is Alice’s payoff and the second is Bob’s). Now consider a game with identical final payoffs but where Bob is initially endowed with nothing. Because the final payoffs are identical, traditional utility theory predicts Alice will pick the same distribution. If there is reference-dependence, however Alice may change her choice – if we consider they payoffs in difference terms (from the reference point) they are (+3,0) or (+5,-5) for the first decision where Bob starts with £10 and (+3, +10) or (+5,+5) for the second where Bob has nothing to begin with. Such a preference reversal would be evidence that traditional models of social preferences, lacking reference dependence, do not tell the full story.
This is definitely an interesting area for future research, and for a more technical discussion you can download a short paper I wrote on the subject as part of my MSc.
Posted in Economics.
– August 29, 2009
For my dissertation I’m using an unbalanced panel, with some singleton observations (cross-sectional units with only one time observation). The fixed-effects estimator I’m using will ignore such units (if there’s only one time period there’s not within effect to measure), however I wanted a way of removing them in Stata before carrying out summary statistics. Oddly enough there doesn’t seem to be anything in the help files or online about doing this, but a quick look at the sourcecode of xtreg2 led me to use the following:
by PANELVAR: gen T_i = _N if _n==_N drop if T_i == 1 drop T_i
Replace PANELVAR with the variable that identifies your cross-sectional units. This is a clever use of the highly useful “by” prefix. This means that the first line is ran across lots of little datasets all with the same PANELVAR. The variable T_i is set to equal _N (the number of observations in each little dataset – i.e. the number of time-periods this particular cross-sectional unit appears in) for the final time period. As a result, any singleton observations will have T_i equal to 1 and can be dropped. This just shows how many insights you can get from examining other people’s commands; the nature of Stata really encourages this type of openness.
BTW I’m going to try to configure the Wordpress install so the more technical posts like this can be easily separated from the general-interest economic articles – please leave a comment if you can think of the neatest way of doing this!
Posted in Economics.
– August 23, 2009
The events of the past couple of years have hardly helped the reputation of the world’s economists, and have led to a great deal of soul-searching within the profession. Following a conference at the LSE last year, the Queen asked a very simple question – “what went wrong?” A group of eminent economists answered with a letter a few weeks ago, and a few more added their own take on things yesterday.
The first letter mentions the underlying imbalances in areas such as the housing and credit markets, and the way in which relatively benign macroeconomic indicators led to a fairly loose monetary policy – central bank’s models didn’t predict any great trouble ahead, so there was no concrete evidence upon which to take pre-emptive action. Their central thesis, however, is that the crisis was not foreseen due to the failure of economists to think about the economic system as a whole:
Everyone seemed to be doing their own job properly on its own merit. And according to standard measures of success, they were often doing it well. The failure as to see how collectively this added up to a series of interconnected imbalances over which no single authority had jurisdiction. This, combined with the psychology of herding and the mantra of financial and policy gurus, lead to a dangerous recipe. Individual risks may rightly have been viewed as small, but the risk to the system as a whole was vast.
This seems to be a reasonable argument. If one looks at the baseline DSGE1framework used by lots of central banks, there are all sorts of implicit assumptions as to how various asset and financial markets work perfectly. Blanchard’s The State of Macro, written just as the financial crisis went critical, acknowledges these sorts of deficiencies in macro models:
The current financial crisis makes it clear that the arbitrage approach to the determination of the term structure of interest rates and asset prices implicit in the basic NK [New Keynesian] model falls short of the mark: Financial institutions matter, and shocks to their capital or liquidity position appear to have potentially large macroeconomic effects.
The second royal letter takes a much more radical tone, arguing that the emphasis on mathematical modelling by economists has been to the detriment of a broader view incorporating insights from psychology, economic history, and how individuals and organisations actually act (rather than what rational agents would do). They write:
… [the previous letter] overlooks the part that many leading economists have had in turning economics into a discipline that is detached from the real world, and in promoting unrealistic assumptions that have helped to sustain an uncritical view of how markets operate.
One of the authors of the letter was interviewed on the Today programme yesterday, and it seems that their criticisms aren’t perhaps the use of maths per-se, but the countless simplifying assumptions that must be made in these models (e.g. about individual behaviour or ignoring out-of-equilibrium behaviour) and are then seemingly ignored when it comes to applying things to real life. This I agree with – whenever you read an economic model it’s vital to continually test the assumptions against real-life and, if they don’t hold, at least have some idea what that means for the model.
That’s the way that the Bank of England works: their BEQM model has a DSGE “core” which is, by necessity, a caricature of reality. However the Bank realises the deficiencies of the model, and so the predictions it gives are supplemented with a number of more ad-hoc corrections that take into account areas ignored by the core model. Sure, it would be better for the Bank to have a model which takes everything into account, but that’s simply not realistic. At least by having a rigorous core to the model (which, unlike other forms of forecasting such as using VARs2, is based on theory and can give structure to why the model predicts something) the bank can back-up their forecasts with hard data. If the model doesn’t fit the real world, then it’s possible to look at what failed and bring the model more into line with the reality – as with any scientific model, BEQM and its cousins provide predictions which can be falsified.
What I disagree with is the notion that mathematical modelling encourages economists to make stupid assumptions and disregard them. Nothing could be further from the truth. Any model, whether it’s defined in mathematical terms (like BEQM) or in more verbose language is a simplification of real-life: that’s the definition of a model. All models therefore make assumptions. At least by using maths, I’m forced to make my assumptions explicit otherwise the results just won’t appear. If, instead, I attempt to describe a model by pure intuition alone, there are all sorts of hidden assumptions that are often hard to point out.
Far from making models detached from the real world, mathematical modelling helps verify just how our simplified version of the world differs from the actual thing. It’s then the job of the economist applying these models to carefully note what assumptions are broken, and either fix things up or at least admit that there’s a possibility that things might not come out entirely as predicted. “Is my financial model based entirely on rational agents? Let’s check what happens if we add some behavioural assumptions instead – do things change much?” These are the sorts of questions that economists can (and do) ask. While it’s certainly true that many practitioners cling dogmatically to false assumptions of frictionless markets and the like, it’s the mathematics itself that highlights these simplifications.
By all means criticise the unwarranted assumptions of mathematical models and the way the don’t see the “big picture”. Just don’t forget that a well-written analytical model is far more truthful about its assumptions than any paper based on hunches and intuitions will be. Caveat emptor…
Posted in Economics.
– August 16, 2009
As something of technophile, I appreciate the utility that people get from Twitter – while I don’t use it myself, microblogging certainly has its place. Unfortunately, the BBC seems to want to elevate Twitter to a status it really doesn’t merit. Over the past year or so, it has become something of a farcical game to see how many mentions there are of Twitter on the front page. The tweet that broke the camel’s back was the one below, seen on the front page this morning:
This story is about the ongoing row in the States and over here about whether the NHS model of healthcare is something to be aspired to. It’s true that part of this story relates to the way in which lots of UK citizens (and the PM) have rallied to the NHS’s defence over Twitter. However, the BBC article in question leads on a completely different angle: David Cameron’s disagreement with a Tory MEP, and his alleged commitment to the NHS. Sure, Twitter is mentioned later, but the most important part of this is the stuff about Cameron, none of which involves Twitter in the slightest. The BBC’s caption by the headline doesn’t mention Twitter, so why does the image? Wouldn’t a stock photo of DC be better?
This is, of course, a terribly minor issue, and it wouldn’t matter were it not for the fact this sort of thing is occurring daily. Every story seems to have quotes from the Twittersphere. It’s a cheap way of getting the standard Vox Pop interviews that they’ve been doing for years, but considering how few people in the country actively use Twitter it seems like it’ll be selecting a highly unrepresentative sample.
No doubt they’re doing this to “get down with the kids”, but it seems to be doing more to alienate people. At least someone at the Beeb has the right idea though – half an hour after it was put up, the image on the front page (the one here is a mockup as I didn’t save it before) was changed to a gurning David Cameron.
Posted in News/Politics, Tech.
– August 14, 2009
The BBC’s report about Ofcom’s recent study into broadband speed contains the following gem:
The Ofcom data comes from 60 million speed tests done using modified routers in 1,600 houses. Many people will probably look at this table and just go “eh?” Surely if they can say that one ISP is faster than another, they could just put it in a simple league table? It’s not quite that simple though, and looking carefully at the table shows quite a few oddities: For example, Tiscali is apparently slower than BT while AOL is not slower than BT – presumably AOL must therefore be faster than Tiscali, but that’s not what the table says.
The answer is to be found in the original Ofcom study. Ofcom did several tests (e.g. peak-time download speeds) but the BBC data is from “Average download throughput speeds, 24 hours, April 2009″ (page 59 of the PDF). There’s a nice little graph of the results there:
This shows us 95% confidence intervals of the average download speed for each ISP (actually the report is slightly confusing – a chart like this could also show the 95% confidence interval within with individual speed test ratings lie, but from the context and the relatively small intervals this doesn’t seem to be the case). What these bars tell us is that “by looking at a small sample of their customers, we’re 95% sure that the average speeds of all AOL’s customers is between about 3.3Mbit/s and 3.9Mbit/s”. It would have been nice for them to have marked the point-estimate of the average, but we know (from the central limit theorem) that the confidence intervals for the mean are approximately symmetrical and that’s probably the approximation they’ve used anyway – the mean is in the centre of the bars.
The next table in the report is the same as the BBC’s. What it’s actually doing is running hypothesis tests along the lines of “Is the average speed of Virgin media likely to be faster than the average speed of AOL?” These are all at the 5% significance level – the table only says “faster” if, from the data we’ve got, there’s a 95% chance that Virgin is faster. That explains what must have mystified a lot of BBC News readers: lots of comparisons between ISPs (e.g. is Sky faster than O2?) are omitted because there’s so much uncertainty. Now it’s too much to ask the BBC to explain all of this in the article, but maybe they could at least have linked to Michael Blastland’s great primer on the subject?
Is there a better way to present the data? There are a few other forms I can think of: for example a matrix with all ISPs on the left and the top, and cells that indicate whether the left ISP is faster, slower or “hard to say” compared to the top ISP. If anything, though, that would take even more explanation. At the very least, the blurb at the top of the table should have said something like this (but a little better worded!):
“As the measured speeds varied so much, it is not always possible to calculate that one provider is faster than another on average. The table below shows comparisons where one ISP was faster in so many tests that their average speed is almost certainly faster”
Personally, I think that the graph above, with a little bit of explanation, would be better than the table. It’s far less cluttered and yet lets you see the actual differences at a glance, providing more information.
Finally, is there not a better benchmark for ISP speed than averages (plus associated confidence intervals)? Some of the variance is probably caused by things like location; while the Ofcom study does take into account the distance from the exchange for ADSL, perhaps certain exchanges are inherently faster for all ISPs due to newer cabling or the like. If this is the case, a more in-depth study might be able to remove such effects, maybe through a regression. What we’re interested in is not “which ISP is faster on average”, but the more cumbersome “which ISP is faster on average, given that we’re comparing across the same area”.
Such an analysis would undoubtedly still have some fairly wide confidence intervals, however. What the public (and the government) would really like is a simple one-dimensional ranking of all ISPs. They could take the average (without the confidence intervals) but this is severely misleading as ISPs may be listed as being faster on average simply due to statistical error. More fundamentally perhaps, a single indicator like “average speed” will miss out many other things that an ISP should be doing like reliability of service or customer support. If the government focussed on just one number then this could actually make things worse in these other areas.
– July 28, 2009
Had to do a bit of detective work today to get to the bottom of an irregularity in the panel I’m using for my dissertation (more about that when I’ve got some actual results!) The dataset itself is an unbalanced panel with organisation-level data on 500 charities for each of the five years from 2003 to 2007. It’s based on the Charities Aid Foundation top 500 fundraising charities lists, which amazingly don’t seem to be available electronically (typing and verifying 60,000 numbers is hard work!). One of the key variables for me is public sector grants. As part of the quick eyeballing I did in OpenOffice, I did a quick graph of this:
Notice anything odd? The years up to 2006 are expected – gradually increasing grants each year due to inflation (the charities are lined up by size hence the rough sawtooth pattern). But the government suddenly seems to get stingy in 2007 with far fewer and smaller grants.
A quick call to the Office of the Third Sector confirmed my suspicions that government spending on charities has been level/rising for a long time (note that the financial data is lagged a couple of years behind the date of the list being compiled, so we cannot blame the credit crunch). This suggests that there’s been some sort of change in charity accounting; without taking this into account or simply purging 2007 from the dataset my regression results could be severely biased.
My first thought was to compare the 2006 and 2007 data to see what had changed – had the public money obviously been reallocated to another variable? It’s hard to come up with a definitive answer this way, however, as all the variables will change considerably from year to year. I accidently stumbled across a better way while importing the data into Stata. I’d attempted to declare the data as a panel, with financial year as the time variable (this is the year the charity did its accounts and not the date of list compilation) and got a complaint about duplicate entries. Following StataCorp’s recommendations, it was easy to isolate the problem: Some charities had evidently submitted their accounts a little late one year, and so the Charities Aid Foundation had to work off the same accounts that they had used for the year before. In particular, there are a few charities where their entries in both the 2006 and 2007 lists are based on accounts for the 05/06 financial year.
As you’d expect, most of the variables for these charities stay the same in 2006 and 2007 – they’re based off the same accounts, after all. Almost invariably, however, income that was under “Public Grants” in 2006 seemed to have been moved to “Incoming Resources From Charitable Activities”, explaining the apparent drop in grants that I’d seen. It appears that the CAF had split the same accounting data into different categories in 2007 (oddly, a few charities had a different “Total Income” figure, which I cannot explain).
What’s the reason for this? I’m waiting for confirmation from CAF, but I’m sure it’s about whether government grants are unconditional or conditional on the charity carrying out various activities (thanks to the NCVO for pointing me in this direction). CAF appears to have made the change in 2007 to stay in line with the Charity Commission’s Statement of Recommended Practice, which states that conditional grants (like the government paying the National Trust to conserve a piece of land) belong under “Incoming Resources From Charitable Activities” while unconditional grants are effectively voluntary donations. Incidently, it turns out that it’s becoming more and more common for government grants to be tied in this way, effectively contracting out parts of the government’s role to the third sector. Whether this is a good thing for charities is debatable.
There are a few lessons that can be taken from this. First of all, anyone constructing a panel dataset will often have to contend with changes of definitions across time – ignore them at your peril! I was able to identify the problem from a glance at the data before it even went into STATA, but narrowing down exactly what had changed in the definitions took several phonecalls, a bit of web searching and some playing with the data. In particular, the way that accounts for a single financial year were repeated across the yearly lists (which initially appeared to do nothing more than limit my dataset) actually provided a neat way to separate real changes in charities’ finances and changes in the way the data had been aggregated. By working out exactly where the missing public grants went, I might now be able to run a few auxiliary regressions to back out a decent estimate of public grants from the “Incoming Resources…” variable (which has them mixed in with other sources of income) and use the 2007 data in the panel.
p.s. I realise this post has been pretty dry, consider it a complementary good to the one below 
Posted in Economics.
– July 27, 2009
The first of a (hopefully occasional) series.
Posted in Economics.
– July 8, 2009
It’s time to shine the spotlight on one of my favourite (ha!) technologies – DRM (Digital Rights Management). At some point, I’m hoping to look into the interesting underlying economics but for now a quick survey of the technological limitations will suffice. Most of these thoughts have been expressed more eloquently by other people, however I feel it’s still important to claim them as my own.
DRM, obviously, covers digital media such as music, videos and computer software. The term “Digital Rights Management” is something of a marketing ploy, however. We all have certain fair-use-type rights, such as making a backup copy, and standard digital formats allow one to do so very easily. DRM is all about imposing additional restrictions on the consumer (e.g. “You cannot copy this”, “You cannot play this video on this hardware”, “You cannot watch this in this country”) at the whim of the manufacturer. When you buy a DRMed file, you’re actually getting a highly specific license that strictly limits what you can do with the file.
Ask a media company why they bother with DRM, and the first answer you’ll probably get will be about defeating piracy. Sounds reasonable – stop people copying the file and it cannot be pirated. Shame there’s no way it can possibly work:
Fact 1: Until images and sound can be streamed directly to your brain, there’s always an “analogue hole” that unencrypted data has to pass through – for example the headphone output on a computer. Sure, you lose a tiny amount of quality going from digital to analogue to back again, but done with the right equipment it’ll be imperceptible to anyone who’s happy with compressed media in the first place. Videos are harder, although someone with enough resources could always find a way to directly grab the signals sent to the individuals pixels of a TFT display. None of this has been necessary so far however, as every known DRM system has been cracked.
Fact 2: It’s trivial to create lossless copies of non-DRMed digital media. So even if your protection works against 99.999% of users, it only takes one to crack it, distribute the file, and make it available to everyone. It’s curious how some people just don’t get this; an article from an industry body insists that partially working copy-protection somehow leads to less free-riding than none at all, but once the file is cracked once then there’s no barrier left to free-riding!
Fact 3: DRMed files always provide less to the consumer than non-DRMed equivalents. If I buy a blu-ray, I’ll have issues trying to play it on my GNU/Linux desktop, transferring it to a portable device, or playing it on my older TFT monitor that doesn’t support encrypted connections via HDMI. End result: The restrictions imposed by Big Media simply make illegal downloads even more attractive!
There are a few caveats to bear in mind though, especially the costs of breaking DRM which vary across media:
Finally, while copying of DRM-stripped files may well be technically easy, legal barriers may prevent people doing so. This doesn’t detract from the argument against DRM as such laws are as hard to enforce whether or not files are originally protected. Note that the DCMA (Digital Millenium Copyright Act) in the States actually makes it an offence to circumvent DRM. This is a terrible precedent as it effectively allows the content creators to make any type of “fair use” an illegal activity. Surely it’s far better for the government to just regulate the activities they consider wrong, rather than delegating this to the record and movie companies?
Posted in Tech.
– July 6, 2009
