I just this morning finished Michael Spence's seminal 1973 article entitled "Job Market Signaling [1]." I'm still musing on it, as it has a lot to chew through. Here are some early comments.
Firstly, Spence introduced the term signaling, but he explicitly didn't define it [2]. I say this to lead into some later remarks. His view was that signaling was something that was undertaken only infrequently; he was specifically looking at the case where the signaler did not acquire the ability to signal well.
Next. The asymmetric information school - which may or may not claim to incorporate signaling - assumes that there is an asymmetry of information, and thus the task is to incentivise the sharing of that information; to whit, reduce the asymmetry and thus make the allocative decision more efficient.
That's not what Spence and Job market signaling is about. Spence explicitly accepts that the market in jobs is symmetric and insufficient; something I had noticed and developed more strongly in my (draft) market for silver bullets. That is, the task here is not to get the individual to reveal information that he holds to his advantage, but to predict something that is otherwise only found out at extreme cost (risk investment in employment decisions).
Which means, amongst other things, that I now have to rewrite my silver bullets paper to take into account that I'm 32 years behind Spence on this point. Lucky it wasn't 33 years, is all I can say. Also luckily for me, his market in education mirrors my market in silver bullets, which leads to the next point: The equilibria in this market arises without reference to the original import of the signal. My model was based on herding, his is based on confirmatory feedback (perhaps like Senge or even Boyd) [3]. The two sit side by side, which means I can build on his and incorporate the two together. I still have a chance of a paper, then!
One point is widely understood; the signal must be expensive for one group and cheap for another. If the costs of acquiring the education are the same for all, there is no value in the signal. This might mean that there is a desire to make sure this is not the case; but this search for apparent differentiation is countered by the feedback equilibria being reached without resort as above.
Finally, Spence actually suggests that markets based on signaling are inefficient, and the signals themselves are not especially correlated with productivity. If his implicit unwritten definition is accepted, signaling is not a good to be pursued, rather a bad to be avoided. That is, the question for the job market, and the education market, is how to avoid the product of education being reduced to the 'bad' of a signal.
This was a surprise for me. I had simply assumed that signals were positive things. Perhaps it is the literature that suggests this, or perhaps it is the crossover to Akerlof, where the lemons market signals positively. This underscores the dictum of going back to the source. Secondary references such as Wikipedia and the Nobel site just don't bring this out.
To underscore this, the paper shows that in some equilibria, it is reasonable to postulate that all parties are strictly worse off in the presence of stable signaling. Further, indices - those signals that cannot be changed and are assumed a priori irrelevant - can create the same equilibria.
Now, that makes sense. That's precisely what I've suggested with some of the much vaunted products that masquerade as security, which is what got us started on this whole signaling kick in the first place. And, if we can recognise that the market for security is one of signaling, and signals are an inferior allocative mechanism, then at least we are some way along in finding ways to deal with that.
[1] Michael Spence, "Job Market Signalling," Quarterly Journal of Economics, v 87(3), 355-374.
[2] It turns out that there are two spellings for signalling or signaling. I haven't as yet worked out the distinction, but I suspect another American English difference here. Here I'll try out his spelling rather than the English I was brough up with.
[3] Spence refers to Myrdal's vicious cycles, which Google puts at a 1957 paper.
Posted by iang at April 17, 2005 03:10 PM | TrackBackMy take on fake security is that it might manage to trick the adversary into deciding that the attack is not worth pursuing. If fake security is substantially cheaper than real security, it might still make sense.
http://www.mast.queensu.ca/~nagydani/missdef.pdf
The above reasoning is applied to missile defenses, but basically any defensive measure would do; if the adversary misjudges our defensive capabilities, it would mount an inefficient attack, or even decide against attacking us at all.
Interesting analysis. It's been a few years since I read Spence for class. I'm curious about a few of your observations:
>the task here is not to get the individual to reveal information that he holds to
>his advantage, but to predict something that is otherwise only found out at
>extreme cost
I understand the conceptual distinction, but can you give me an example of a situation in which the latter holds, but the former is not applicable?
>this search for apparent differentiation is countered by the feedback equilibria
>being reached without resort as above
Could you clarify what you mean by this?
>markets based on signaling are inefficient, and the signals themselves are not
>especially correlated with productivity
Doesn't the innefficiency of signaling follow directly from the lack of inherent value in the signal? As I recall, this was an obvious, minor point: if a market necessitates an added cost (education) that does not directly increase value, education is "innefficient." If we start arguing that school improves skills to make better workers, or [bigger picture] keeps younger workers out of the labor force, (or just employs professors who would otherwise drag the economy :) ) then the social equilibria shift, no?
Also, do you have a circulating draft of your paper?
Allan, your first question - Spence's original market was just that: the job market. the task that the employer has is to work out whether a prospective employee is likely to be productive; if we assume that all candidates more or less present themselves well in an interview, we still have a problem in translating that to productivity. In practice the only reliable way is to road test the employee, either on a short term contract or in a "realistic test" scenario. The latter is only a mild predictor, and the former is spoofable ("best behaviour"). In the end, the employee has to invest 1-3 months worth of time (and costs) to find out, which means that it truley is a risky investment.
Posted by: Iang at April 19, 2005 04:21 AMAllan,
>this search for apparent differentiation is countered by the feedback equilibria
>being reached without resort as above
my original thoughts here were along the lines of, well, if Spence is saying that a good signal is one that is expensive for some and cheap for others, then the task is to select such signals. Consider an MBA - they are quite expensive, but in the US and moving out, it is now possible to do much cheaper MBAs by correspondence, etc. Lowering the cost of the MBA lowers the value of the signal, so this might be considered a bad thing.
Yet, Spence is also saying that stability is reached in the market due to a feedback loop that only shifts on disconfirming information. That is, unless decidedly strong data turns up year after year that says the MBA does not correlate with productivity, then an equilibria is reached on that signal, and it remains a strong one. But, and this is the crucial point, this equilibria is not based on whether the MBA leads to productivity - it might have arisen for other reasons and simply stuck there.
Literally, what Spence is saying is that signals are those things that don't necessarily tell us about productivity directly, but are stuck in a feedback loop as if they tell us about productivity.
So getting back to the first point, there's not much point in working to improve that signal (by for example making sure that an MBA is more expensive). What we really need to do is more subtle; perhaps we should disrupt the feedback loop? Perhaps we need to inject some data, or create new feedback loops? I don't know, and I'm minded to John Boyd's "inside own loop" syndrome here, where he concludes that once an organisation is in this loop, there is no known way to get out.
Posted by: Iang at April 19, 2005 04:30 AM> Doesn't the innefficiency of signaling follow directly from the lack of inherent value in the signal? As I recall, this was an obvious, minor point: if a market necessitates an added cost (education) that does not directly increase value, education is "innefficient."...
Right, exactly. However one interprets it. But curiously enough, I'm coming at it from a security and net-sources point of view and I've not picked up on that point. To you it might seem a minor point; to me, I had assumed from things people said (!!) that signals were good things. This is a complete change!
Also, I think it overshadows his real point which is that in an insufficient information market, something arises anyway. This is a point I try to bring out in my paper: what and why is that something? how do we identify this, and what can we do about it?
> Also, do you have a circulating draft of your paper?
It's at http://iang.org/papers/market_for_silver_bullets.html
It is at a form where it is cohesive, I just have to rewrite all the parts relating to the herding and so forth to account for Spence's paper. So it is likely to change quite dramatically in detail. I knew in my heart that the things I'd discovered had to have been addressed before so all I'm doing is playing catchup again :)
I also have on my reading list Rothschild&Stiglitz and Vickrey. Unfortunately these papers are not on the net, and I have to trip along to the library to get at them.
Posted by: Iang at April 20, 2005 10:08 AM