It’s Hard Being a Bear (Part Two)

Flattr this!

One of the rea­sons I’m still a bear on the econ­omy is because the econ­o­mists in the opti­mists camp are rely­ing upon very bad eco­nomic the­ory. If that the­ory is telling them good times are ahead, that’s one of the best pre­dic­tors of bad times you could have.

This isn’t because the opti­mists are bad econ­o­mists, bad peo­ple, or any other per­mu­ta­tion: most econ­o­mists I know are good at what they do, and are very well inten­tioned too.

It’s just that they were taught a crock of non­sense at uni­ver­sity, and they now build mod­els based on a crock of non­sense that they erro­neously believe to be accu­rate descrip­tions of the real world.

There are so many bits of non­sense in eco­nomic the­ory that it would take a book to detail them all, but com­mon to many of them is the fol­low­ing dilemma:

Almost every­thing econ­o­mists believe is pos­si­bly true at the level of an iso­lated indi­vid­ual, but almost cer­tainly false at the level of an econ­omy.

The most egre­gious exam­ple of this is one the­ory that even most econ­o­mists are now will­ing to admit is false: the “Cap­i­tal Assets Pric­ing Model”, which preached that stock mar­ket price shares accu­rately, that the amount of debt finance a com­pany has doesn’t affect its value, and many other notions that have gone up in smoke dur­ing the GFC.

The CAPM is actu­ally derived from a model of the behav­iour of an iso­lated stock mar­ket investor. The investor has expec­ta­tions about how all the shares in the mar­ket are going to per­form in the future, and the abil­ity to lend or bor­row money at a “risk free” rate. She then com­bines a port­fo­lio of shares that gives her the best risk-return trade­off with this risk-free asset—borrowing money to lever her invest­ment in the mar­ket if she’s a “risk-seeker”, and lend­ing money if she’s con­ser­v­a­tive.

William Sharpe, the devel­oper of the model, was then stuck with a dilemma: how to go from a model of a sin­gle, iso­lated investor, to one of the entire stock mar­ket? He did what so many neo­clas­si­cal econ­o­mists before him had done: he “assumed a mir­a­cle”. To quote Sharpe:

In order to derive con­di­tions for equi­lib­rium in the cap­i­tal mar­ket we invoke two assump­tions:

First, we assume a com­mon pure rate of inter­est, with all investors able to bor­row or lend funds on equal terms.

Sec­ond, we assume homo­gene­ity of investor expec­ta­tions: investors are assumed to agree on the prospects of var­i­ous investments—the expected val­ues, stan­dard devi­a­tions and cor­re­la­tion coef­fi­cients.

Need­less to say, these are highly restric­tive and undoubt­edly unre­al­is­tic assump­tions… (Sharpe 1964, pp. 433–434)”

Sharpe thus went from a fea­si­ble the­ory of a sin­gle investor to a ludi­crous the­ory of the entire mar­ket, by assum­ing (a) that all investors are iden­ti­cal (except for their atti­tudes toward risk) and (b) that all investors can accu­rately pre­dict the future.

Though he didn’t admit point (b) in this paper, it was made explicit by one-time believ­ers, Eugene Fama and Ken French, in a later exam­i­na­tion of the model’s empir­i­cal fail­ure. They noted that the mad assump­tions could be why it had failed:

The first assump­tion is … investors agree on the joint dis­tri­b­u­tion of asset returns … And this dis­tri­b­u­tion is the true one—that is, it is the dis­tri­b­u­tion from which the returns we use to test the model are drawn.” Fama and French (2004, p. 26)

So for four decades, econ­o­mists applied a the­ory of the stock mar­ket that was based on the absurd assump­tion that every last stock mar­ket investor is a Nos­tradamus: all investors agree about the future and their expec­ta­tions about the future are cor­rect.

If only this were an iso­lated piece of non­sense. Unfor­tu­nately, it’s indica­tive of a fail­ing that is endemic to con­ven­tional “neo­clas­si­cal” eco­nomic the­ory. They develop a model which starts from an iso­lated “Robin­son Cru­soe” indi­vid­ual; then when they bring in “Man Fri­day”, they pre­tend that rela­tions between indi­vid­u­als don’t alter the story in any sig­nif­i­cant way.

Unfor­tu­nately, they do. So the indi­vid­ual para­bles with which econ­o­mists regale us, and which make sense on an indi­vid­ual scale, don’t apply at the aggre­gate level.

Macro­eco­nomic the­ory, which is the real focus of inter­est now that the GFC has brought “The Great Mod­er­a­tion” to a close, is just as bad. For decades it has been dom­i­nated by what is known as the IS-LM model, which most econ­o­mists believe was devel­oped by Keynes.

It wasn’t. Its orig­i­nal author was John Hicks, a con­ser­v­a­tive oppo­nent of Keynes’s at the time, and he devel­oped the model as a means to inter­pret Keynes from a neo­clas­si­cal point of view. The model emas­cu­lated what was orig­i­nal in Keynes’s Gen­eral The­ory, and this bowd­lerised ver­sion of Keynes was then demol­ished by Fried­man in the 1970s to usher in the Mon­e­tarist phase.

Mon­e­tarism has come and gone, but the IS-LM model still under­pins most of the mod­els used by Trea­suries and Cen­tral Banks around the world.

Which is a curi­ous thing, since one per­son who argued emphat­i­cally that this model should be aban­doned was… John Hicks. As so often hap­pens in eco­nom­ics, a “young turk” found wis­dom in his old age, and realised that his model was unten­able.

Like so many mod­els in eco­nom­ics, the IS-LM model starts as a model with two inter­sect­ing lines. The axes of the dia­gram are income (on the hor­i­zon­tal) and the rate of inter­est (on the ver­ti­cal).

The IS curve, which slopes down­wards, pur­ports to show all com­bi­na­tions of the level of out­put and the rate of inter­est that make sup­ply equal to demand in the goods mar­ket. The LM curve, which slopes upwards, shows all com­bi­na­tions of out­put and the rate of inter­est that make sup­ply equal demand in the money mar­ket. The inter­sec­tion of the two curve shows where both the goods and money mar­ket are in equi­lib­rium.

At uni­ver­sity, econ­o­mists are taught to con­sider what might hap­pen when, for instance, the actual com­bi­na­tion of the rate of inter­est and the level of out­put are such that there is excess sup­ply in the goods mar­ket and excess demand in the money mar­ket, and so on. The macro­eco­nomic mod­els they build after leav­ing uni­ver­sity are then explic­itly based on the IS-LM frame­work, with the assump­tion that the econ­omy will always tend towards the point where they two curve inter­sect.

One issue might be obvi­ous to astute observers: how can you model the macro­econ­omy with­out hav­ing an explicit model of the labour mar­ket? Hicks omit­ted this mar­ket on the basis of the neo­clas­si­cal assump­tion that, in a 3 mar­ket world, if two of the mar­kets are in equilibrium—money and goods—then the third also has to be in equi­lib­rium. But this assumption—which I can crit­i­cise on its own grounds—can’t be applied when there is dis­e­qui­lib­rium. If the econ­omy is “on” it’s IS curve, so that the goods mar­ket is in equi­lib­rium, but “off” its LM curve, so that the goods mar­ket is out of equi­lib­rium, then the third “miss­ing” mar­ket must also be in dis­e­qui­lib­rium.

So the IS-LM model is only valid if the econ­omy is in equilibrium—at which point, of course, there’s no role for pol­icy: “if it ain’t in dis­e­qui­lib­rium, don’t fix it” (econ­o­mists also waf­fle on about “Key­ne­sian” and “Clas­si­cal” loca­tions for the IS curve, but I’ll ignore that pseudo-debate here).

After repeated dis­cus­sions with non-ortho­dox economists—especially Paul David­son, the edi­tor of the Jour­nal of Post Key­ne­sian Economics—John Hicks came to appre­ci­ate this point, and in 1979 explic­itly rejected the model:

I accord­ingly con­clude that the only way in which IS-LM analy­sis use­fully survives—as any­thing more than a class­room gad­get, to be super­seded, later on, by some­thing better—is in appli­ca­tion to a par­tic­u­lar class of causal analy­sis, where the use of equi­lib­rium meth­ods… is not inap­pro­pri­ate… [but] When one turns to ques­tions of pol­icy, the use of equi­lib­rium meth­ods is still more sus­pect…” Hicks (1981, pp. 152–153)

So the father of IS-LM analy­sis jus­ti­fi­ably dis­owned his child three decades ago—and yet the teach­ing of macro­eco­nom­ics still cen­tres on this model, and it forms the core of most neo­clas­si­cal mod­els of the macro­econ­omy.

It’s now being super­seded by mod­els that are sup­posed to have “good micro­eco­nomic foundations”—the so-called Dynamic Sto­chas­tic Gen­eral Equi­lib­rium (DGSE) mod­els. The authors of these mod­els do not know that the “good micro­eco­nomic foun­da­tions” on which they base their mod­els have also been shown to be false!

Here the prob­lem again relates to aggre­ga­tion: the model of an iso­lated indi­vid­ual makes it easy to derive a demand curve for that indi­vid­ual. But when you try to derive a demand curve for a mar­ket, you have the dilemma that chang­ing prices also changes incomes. Neo­clas­si­cal econ­o­mists showed that a mar­ket demand curve could wob­ble all over the place, even if the demand curves of every indi­vid­ual in it had the stan­dard “down­ward slop­ing” shape.

This is stated emphat­i­cally and clearly in a “bible” of neo­clas­si­cal eco­nom­ics, the Hand­book of Math­e­mat­i­cal Eco­nom­ics:

mar­ket demand func­tions need not sat­isfy in any way the clas­si­cal restric­tions which char­ac­ter­ize con­sumer demand func­tions…

The impor­tance of the above results is clear: strong restric­tions are needed in order to jus­tify the hypoth­e­sis that a mar­ket demand func­tion has the char­ac­ter­is­tics of a con­sumer demand func­tion.

Only in spe­cial cases can an econ­omy be expected to act as an ‘ide­al­ized con­sumer’.” Shafer and Son­nen­schein (Hand­book of Math­e­mat­i­cal Eco­nom­ics, 1993, p. 672)

Yet DGSE mod­els rep­re­sent the house­hold sec­tor of the econ­omy as a sin­gle, util­ity-max­imis­ing indi­vid­ual! Why? Because their authors believe that it’s quite OK to model the entire econ­omy as a sin­gle indi­vid­ual. Why? Because the text­book from which they learnt their eco­nom­ics told them so. This is now one of the stan­dard texts for Hon­ours and PhD edu­ca­tion in eco­nom­ics puts the

Unfor­tu­nately … The aggre­gate demand func­tion will in gen­eral pos­sess no inter­est­ing prop­er­ties … The neo­clas­si­cal the­ory of the con­sumer places no restric­tions on aggre­gate behav­iour in gen­eral.” (Var­ian 1992)

But then he states that this prob­lem can be avoided by assum­ing that

all indi­vid­ual con­sumers’ indi­rect util­ity func­tions take the Gor­man form… [where] … the mar­ginal propen­sity to con­sume good j is inde­pen­dent of the level of income of any con­sumer and also con­stant across con­sumers… This demand func­tion can in fact be gen­er­ated by a rep­re­sen­ta­tive con­sumer…” Var­ian (1992)

Stripped of its jar­gon, this says that mar­ket demand curves will slope down­wards if we assume that there’s just one indi­vid­ual and just one com­mod­ity! Stated so baldly, no-one could take this argu­ment seriously—let alone base mod­els of the econ­omy on it. But stated in the oblique and sani­tised man­ner of an eco­nom­ics text­book, the absurd becomes the norm.

The edu­ca­tion of econ­o­mists at most uni­ver­si­ties (not, I am pleased to say, my own uni­ver­sity) is there­fore the farce that turns fal­lacy into tragedy. Fatal flaws in the the­ory that are evi­dent in the orig­i­nal research papers of the dis­ci­pline are gloss over in eco­nomic text­books and the stan­dard sub­jects based on them.

There­fore most prac­tic­ing econ­o­mists, who read the text­books but not the orig­i­nal lit­er­a­ture, are com­pletely unaware of the piti­ful foun­da­tions on which their care­fully crafted mod­els are built. Their assur­ances about the future are there­fore utterly unre­li­able.

I’ll hap­pily remain a bear while the­o­ries like neo­clas­si­cal eco­nom­ics pre­dict the immi­nent arrival of spring.

Fama, E. F. and French, K. R. 2004, ‘The Cap­i­tal Asset Pric­ing Model: The­ory and Evi­dence’, The Jour­nal of Eco­nomic Per­spec­tives, vol. 18, no. 3, pp 25–46.

Hicks, J. 1981, ‘IS-LM: An Expla­na­tion’, Jour­nal of Post Key­ne­sian Eco­nom­ics, vol. 3, no. 2, pp 139–154.

Shafer, W. and Son­nen­schein, H. 1993, ‘Mar­ket demand and excess demand func­tions’, Hand­book of Math­e­mat­i­cal Eco­nom­ics, vol. 2, Else­vier.

Var­ian, H. R. 1992, Micro­eco­nomic analy­sis, W.W. Nor­ton, New York.

Bookmark the permalink.
  • Steve,

    Unfor­tu­nately I am not at the moment work­ing on these mod­els but they come from work I did back in the 1960’s and then later in the 1990’s. In the 1960’s we used the meth­ods to sim­u­late atomic bomb explo­sions, and to model the eco­nomic out­put of a power grid, and in the 1990’s I worked on near­est neigh­bour net­works as a way of nav­i­gat­ing search spaces (but it is fun­da­men­tally the same prob­lem) and use the same tech­niques as mete­o­rol­ogy. In between times I worked on a vari­ety of meth­ods to solve com­pu­ta­tion­ally hard prob­lems using iter­a­tive evo­lu­tion­ary tech­niques such as genetic algo­rithms and Tabu searches. These meth­ods are very good at find­ing near opti­mal solu­tions with many objec­tive func­tions and with mul­ti­ple restric­tions.

    The ideas are expressed very well in the work of Miller and Page where their book “Com­plex Adap­tive Sys­tems” gives an eas­ily acces­si­ble descrip­tion of the approach.

    Over the past ten years I have been involved in con­struct­ing large scal­able money and related sys­tems. We built an oper­a­tional micro­pay­ments sys­tems which was stopped in its tracks by bank­ing reg­u­la­tions that makes it ille­gal for an organ­i­sa­tion to hold money on deposit for gen­eral use unless the organ­i­sa­tion has a bank­ing license .

    In the past five years we have built an elec­tronic iden­ti­fi­ca­tion sys­tem for use by organ­i­sa­tions that have to com­ply with the anti money laun­der­ing reg­u­la­tions. This is turn­ing out to be a com­mer­cial suc­cess and I am hop­ing to have more time to devote to the “money” prob­lem.

    While build­ing these sys­tems I have been struck by the sim­i­lar­i­ties to the evo­lu­tion­ary algo­rithms used to solve dif­fi­cult prob­lems. It then hit me that the eco­nomic sys­tem can be thought of as giant evo­lu­tion­ary algo­rithm which is try­ing to opti­mise “mon­e­tary value”.

    When you start to think about it this way and you realise that you can mon­i­tor what hap­pens in the algo­rithm in real time (because nowa­days it is an elec­tronic sys­tem) and you realise you can change some of the rules of the algo­rithm then eco­nomic mod­el­ling can be viewed in a dif­fer­ent way.

    Mod­els are use­ful because they enable us to pre­dict what goes on by “run­ning sim­u­la­tions”. How­ever, because the eco­nomic sys­tem is so con­nected and wired up we actu­ally do not have build a the­o­ret­i­cal model as we have the capa­bil­ity of con­struct­ing a pretty close replica of the actual sys­tem that can be cal­i­brated by observ­ing the day to day trans­ac­tions.

    It is this build­ing of repli­cas that inter­ests me.

    Very sim­pli­fied mod­els do give us insights into what hap­pens but it is really dif­fi­cult to pre­dict what will be the effect of the stim­u­lus pack­age, or of a tax cut, or of chang­ing the bank­ing reg­u­la­tions in the more com­plex real sys­tem. How­ever, a good enough agent based replica that matches pretty well what goes on would be an excel­lent tool to get some insights into what hap­pens and it would not take too much build — in fact I believe that it would cost a lot less than the crude mod­els that trea­sury cur­rently use and that you briefly describe.

    In sum­mary the approach sug­gested is

    To build an elec­tronic replica of the sub­sec­tions of the exist­ing econ­omy where eco­nomic agents are rep­re­sented and which inter­act in well defined ways. 

    To get an idea of how this may work take a look at the lit­er­a­ture on “the ant opti­mi­sa­tion algo­rithm”.

    In the mean­time I am look­ing at the effect of chang­ing the rules on the issu­ing of loans to build new pro­duc­tive assets. It is my belief that we can change the rules in this area so that the money sup­ply can be var­ied in a more con­trolled man­ner than our cur­rent method of loan­ing money against exist­ing assets includ­ing other loans. Unfor­tu­nately I do not have the time or the resources to model this but I do have the resources to build a real sys­tem with dif­fer­ent rules.