The most damning critique of DSGE Author: Noah Smith If DSGE models work, why dont !eo!le use them to get rich" #hen I studied macroeconomics in grad school, I was told something along these lines: $DSGE models are useful for !olicy ad%ice &ecause they 'ho!efully( !ass the )ucas *ritique+ If all you want to do is forecast the economy, you dont need to !ass the )ucas *ritique, so you dont need a DSGE model+$ This is usually what I hear academic macroeconomists say when asked to e!lain the fact that essentially no one in the !ri%ate sector uses DSGE models+ -ri%ate. sector !eo!le cant set economic !olicy, the argument goes, so they dont need)ucas *ritique.ro&ust models+ The !ro&lem is, this argument is wrong+ If you ha%e a model that &oth A( satis/es the )ucas *ritique and 0( is a decent model of the economy, you can make huge amounts of money+ This is &ecause although any old s!readsheet can &e used to make unconditional forecasts of the economy, you need )ucas.ro&ust models to make goodpolicy-conditional forecasts+ )et me e!lain+ An unconditional forecast is when you say $GD- growth will &e 1+23 net year$, or $in4ation will &e 5+63 net quarter$+ 7or this kind of thing,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
If DSGE models work, why dont !eo!le use them to get rich"
#hen I studied macroeconomics in grad school, I was told something along these
lines:
$DSGE models are useful for !olicy ad%ice &ecause they 'ho!efully( !ass the
)ucas *ritique+ If all you want to do is forecast the economy, you dont need to
!ass the )ucas *ritique, so you dont need a DSGE model+$
This is usually what I hear academic macroeconomists say when asked to e!lain
the fact that essentially no one in the !ri%ate sector uses DSGE models+ -ri%ate.sector !eo!le cant set economic !olicy, the argument goes, so they dont
need)ucas *ritique.ro&ust models+
The !ro&lem is, this argument is wrong+ If you ha%e a model that &oth A( satis/es
the )ucas *ritique and 0( is a decent model of the economy, you can make huge
amounts of money+ This is &ecause although any old s!readsheet can &e used to
make unconditional forecasts of the economy, you need )ucas.ro&ust models to
make good policy-conditional forecasts+
)et me e!lain+ An unconditional forecast is when you say $GD- growth will &e1+23 net year$, or $in4ation will &e 5+63 net quarter$+ 7or this kind of thing,
!eo!le, using their silly s!readsheets with their 68s.%intage -hilli!s *ur%es, will
forecast a rise in GD- growth, so they will !ay a lot for stocks, e!ecting higher
!ro/ts from the increased growth+ 0ut wise DSGE modelers, using the No&el.
winning and ostensi&ly )ucas.ro&ust <ydland.-rescott 5=>1 model, know that the
-hilli!s *ur%e is not structural+ They know that the !romised growth will not
occur, so as soon as stocks &ecome o%er!riced, they short the S?-+ #hen the
ho!ed.for growth does not materiali@e and stocks fall, the DSGE modelers rea! ahuge !ro/t at the e!ense of the s!readsheet modelers+
Now thats a &it of an old eam!le, so lets take a more modern one+ Su!!ose the
7ed launches a new !rogram of E+ *le%er DSGE modelers, armed with Ste%e
#illiamsons 185B E !a!er, know that E will &e de4ationary rather than
in4ationary 'as most !eo!le think(+ This allows them to take other in%estors, who
are armed only with s!readsheets, for a ride, shorting TI-S and &uying Treasuries+
Coila . instant riches+ #illiamson himself endorses this idea, writing:
If it does anything, E will lower the in4ation rate o%er the long run+ And thelong run comes sooner than you might think, i+e+ if E gi%es you a short.run
increase in in4ation, then if its like ty!ical monetary easing, then that eect lasts
only a year or two+ ;ore to the !oint, there are other forces !ost./nancial crisis
that will cause the real interest rate on safe assets to rise, and in4ation to fall
further, so long as the 7ed kee!s short nominal rates at or near the @ero lower
&ound+ And there are good reasons to think that the 7ed will &e stuck at the @ero
lower &ound inde/nitely+ *onclusion: e!ect less in4ation rather than more+ That
has to matter for your portfolio choices. 'em!hasis mine(
So as we see, a )ucas.ro&ust DSGE model has the !otential to make its wielders a
)FT of money+ This is es!ecially true in the current en%ironment, wherecorrelations are high and macro e%ents ha%e &ecome much more im!ortant to
in%estors !erformance+
0ut not necessarily+ 0eing )ucas.ro&ust is necessary for making o!timal !olicy.
contingent forecasts, &ut it is not sufcient + ou also need the model to &e a good
model of the economy+ If your !arameters are all structural, &ut you%e assumed
the wrong microfoundations, then your model will make &ad !redictions e%en
though its )ucas.ro&ust+
So now lets get to the !oint of this !ost+ As far as Im aware, !ri%ate.sector /rmsdont hire anyone to make DSGE models, im!lement DSGE models, or e%en scan
the DSGE literature+ There are a lot of /rms that make macro &ets in the /nance
industry . in%estment &anks, macro hedge funds, &ond funds+ To my knowledge,
none of these /rms s!ends one thin dime on DSGE+ I%e called and emailed
e%eryone I could think of who knows what /nancial.industry macroeconomists do,
and theyre all unanimous . they%e ne%er heard of anyone in /nance using a
DSGE model+
If you know someone who does, !lease re!ly in the comments+ Im sure theres
someone out there+ 0ut e%en if there is, they ha%ent soared to fame and fortune
on the &ack of their DSGE model+
So may&e theyre Hust using the wrong DSGE models" ;ay&e theyre
using#illiamson '1851( instead of #illiamson '185B(+ I mean, after all, there is a
huge, %ast, unending array of DSGE models out there, most of which !ur!ort to
e!lain the entire macroeconomy, and most of which are thus mutually eclusi%e
at any !oint in time+ ;ay&e two or three of them are right at any gi%en !oint intime, &ut may&e this set switches around as conditions change+ -erha!s /nance.
industry !eo!le are sim!le una&le to !ick out the right DSGE model to use on any
gi%en day+
0ut if /nance.industry !eo!le cant know which DSGE model to use, how can
!olicymakers or !olicy ad%isors"
In other words, DSGE models 'not Hust $7reshwater$ models, I mean the whole kit
? ca&oodle( ha%e failed a key test of usefulness+ Their main selling !oint .
satisfying the )ucas critique . should make them %ery %alua&le to industry+ 0utindustry shuns them+
;any economic technologies !ass the industry test+ *om!anies !ay !eo!le lots
of money to use auction theory+ *om!anies !ay !eo!le lots of money to
use %ector autoregressions+ *om!anies !ay !eo!le lots of money to
use matching models+ 0ut com!anies do not, as far as I can tell, !ay !eo!le lots
of money to use DSGE to !redict the eects of go%ernment !olicy+ ;ay&e this will
change someday, &ut its &een B1 years, and no ones touching these things+
As I see it, this is currently the most damning critique of the whole DSGE!aradigm+
!dates:
;att glesias comments+
Tony ates comments, at greater length+
The anonymous deni@ens of EJ;K discuss DSGE and the !ri%ate sector+Interesting !ost:
Insider in a small, &outique !ri%ate equity /rm+ #e do not use DSGE models, &ut
we do use DSGE models as a screening de%ice, in the same s!irit as graduate
!rograms use real analysis+ It turns out that we can get these guys chea!er than
;0As, and they ha%e similar le%els of /re!ower+
So DSGE does function as a kind of signaling . the a&ility to make DSGE models is%alua&le to com!anies e%en if the models themsel%es arent, &ecause they
indicate general intelligence, com!uter skills, creati%ity, andLor work ethic+ That
makes sense+
*hris Mouse, my old /rst.year macro instructor and a !rof at the ni%ersity of
;ichigan, has A( started a &log, which you should read, and 0( attem!ted to
re&ut this !ost+ *heck it out 7un fact: The argument in quotes at the to! of this
!ost is actually !ara!hrased from stu *hris Mouse said to me, which is %ery
similar to what he now writes in his !ost+ I must say, though, I think *hris !ost
actually reiterates and strengthens the !oint I make in my !ost+ The dierence&etween !olicy.conditional and other forecasts is key here+ If the a%erage
!olicymaker can use a DSGE model to im!ro%e her !redictions of the results of
her !olicies, then the a%erage /nancial trader can use the same model to
im!ro%e her !redictions of the results of the !olicymakers !olicies+
Tyler *owen thinks Im too quick to dismiss DSGE, and that instead, critiques like
this one should merely make us $downgrade$ DSGE+ This is a %ery 0ayesian
a!!roach, and I like it+ 0ut notice that 0ayesian &eliefs de!end on your !riors+ If
you started out thinking that DSGE was totally awesome, then critiques like this
one should make you tem!er your enthusiasm+ 0ut if you started out thinkingthat DSGE was hocus.!ocus, then critiques like this should tend to strengthen
your &elief somewhat+
-osted at 5:59 A;
Email This0logThisShare to TwitterShare to 7ace&ookShare to -interest
5B2 comments:
5+
Andrew Lilley5:12 A;
I can con/rm the e!ectation, Australian in%estment &anks and hedge funds do
not use DSGE models+ Cery few who are em!loyed to forecast would know what
0ut++ markets are eOcient so wise DSGE modellers will not make money+ The
DSGE !aradigm wins again
Ke!ly
B+
AnonymousB:1B A;
DSGE models historically ha%e com!letely ignored /nance+ 7inance is sim!ly
re!lying in kind+
Ke!ly
2+
Will 2:28 A;
Arent DSGE models !rimarily used in !olicy institutions in order to !ro%ide a
narrati%e as to what has &een occuring within the economy" ;ainly &ecause at
some!oint economists need to talk to non.economists who desire to ha%e some
form of structuredLcausal e!lanation of why things are the way they are"
Fut of interest what macro models do /nancial institutions use" CAKs"
Ke!ly
9+
Robert 9:8= A;
Sometimes I &ore myself+ In fact when I read $)ucas critique$ I usually &ore
myself+ I will &ore you too+ I fear I cant resist discussing the meaning of $)ucas
ro&ust$+ It is quite dierent from the meaning of $ho!efully )ucas ro&ust$+
#e can go &ack to the source of the )ucas critique to a%oid the word $)ucas$+ That source was named Jaco& ;arshack
htt!:LLen+wiki!edia+orgLwikiLJaco&P;arshak ')ucas cited him(+ Me was also a
minister in a ;enshe%ik go%ernment once+ Me introduced the word $structural+$
Ah thats the ticket+ They !hrase $)ucas.ro&ust$ can &e translated to $your
!arameters are all structural+$
F< so what does $structural$ mean in this contet " It is currently often used to
mean $ha%ing something to do with o!timi@ing$ &ut it originally meant !olicy
in%ariant+ Now it is not necessary to write down a model in which agents o!timi@e
to ha%e !olicy in%ariant !arameters .. the Newtonian model gra%ity gi%es !olicyin%ariant close enough to &e useful e%en if we do not discuss the utility function
Similarly the fact that one writes down a model in which a !arameter a!!ears in
a utility function does not make em!irical estimates of that !arameter !olicy
in%ariant+ that would &e true if the model were reality+ The fact that the model /ts
a%aila&le data F< tells us no more than it would if no techniques of o!timi@ationwere not used in sol%ing the model+ The relationshi! &etween our conditional
forecasts and conditional !ro&a&ility distri&utions cant de!end on whether we or
not we consider the regression coeOcient of consum!tion growth on the real
interest rate the estimate of a !arameter of a utility function or Hust an ad hoc
regression coeOcient+
The argument that models with micro foundations are )ucas.ro&ust makes no
sense at all+ I think it has !recisely eactly @ero merit of any kind+
Ke!ly
Ke!lies
5+
Noah Smith=:B6 A;
Sure+ 0ut I say Qe%en grantingQ )ucas.ro&ustness to a su&set of dsge models,
they fail+the market test+
Reply
R+
Robert 9:51 A;
0y the way, I &ore myself &ut you dont &ore me+ This is a &rilliant !ost+ The
!recise quantitati%e forecasts 'in the conditional sense you descri&ed(, sim!ly
&ecause macro is tough . so there would &e no !oint in constructing a !ri%ate
DSGE model to get !recise E.conditional !ro/t.maimi@ing !ortfolio weights, or
whate%er+
0ut models can still !ro%ide qualitati%e insights . if ha!!ens and assum!tionsA,0,* are a!!roimately true, then will increaseLdecreaseLstay the same+ At the
same time, theres a natural di%ision of la&or . academic economists generate
new insights with DSGE models, industry economists a!!ly these insights '!assed
to them through tet&ooks or sur%ey !a!ers( when analy@ing s!eci/c questions
they face, according to their &est Hudgement a&out which insights are most
rele%ant for their situation+
Do industry !eo!le use modern macroeconomic theory in this way" I dont know+
If they dont, there could &e se%eral reasons . may&e DSGE are worthless+ 0ut in
that case, what kind of theory !eo!le QdoQ use" Fld.style <eynesian models"Intuition" Keciting !ages from ;ises" That would &e interesting to know+
Fn the other hand, may&e it Hust takes time for theory to trickle down from
research frontier to tet&ooks, and !eo!le will use them more in the future+ Fr
may&e most DSGE models dont actually &ring that much new insights . e+g+ the
whole New.<eynesian research !rogram seems to &e mostly a&out deri%ing
standard undergraduate <eynesian macro with intertem!oral o!timi@ation 'which
I think is %alua&le, &ut it wont change !eo!les decision %ery much(+
Ke!lyKe!lies
5+
#nlearnin$econ6:B9 A;
I agree with this . economics should not run &efore it can walk, and e!laining the
general, stylised &eha%iour of the economy is surely the &est starting !oint we
can ho!e for+
F&%iously theres still the issue of whether DSGE does this suOciently+ I think it
still suers from the e.ante choice of model !ro&lem the Noah references here+
1+
%r&ys B:16 -;
The !ro&lem with this argument is that some DSGE models ha%e %ery dierent
qualitati%e !rediction from the market mo%ements+ In other words, sim!le
directional trades should make you money, &ut they dont+
;ore generally the industry doesnt hire new <eynesians, freshwater !eo!le or
market monetarists+ Ftoh they !ay &ig &ucks for ;;T.related analysis ';c*ulley
at -imco, retired rich, Mat@ius at GS(+
Ke!ly
Ke!lies
5+
Won)s Anonymous5:58 -;
I thought Mat@ius was su!!osed to &e rather mainstream in his economics, rather
than ha%ing any association with ;;T+
1+
%r&ys B:1= -;
7unny, Mat@ius endorsed ;; a year and a half ago+
B+
*an)' 2:1= -;
-I;*F ne%er used ;;T in its &ond trading decisions+ ;c*ullley only looselystarted discussing some -ost.<eynesian elements of economics in the latter years
of his career+
And Jan Mat@ius sometimes uses #ynne Godleys sector /nancial &alances+ Me
doesnt $do ;;T$ in any meaningful sense and is actually much more
#ell, a DSGE !ro!onent would sim!ly counter that the industry is run &y a &unch
of herd.following idiots, surely"
Ke!ly
Ke!lies
5+
%r&ys B:22 -;
This comment has been removed by the author.
Reply
59+
Anonymous=:58 A;
I trade with econometric models+ I ha%e tried+ Fr rather, my research associate
'fresh out of a school(, tried hard+ Me is con%inced that DSGE models are su!erior+
0ut &acktesting !ro%ed to &e a rather frustrating eercise for him+ Not that I was
sur!rised+ 0ut sometimes the only way to learn is to fail yourself+
This &rings u! another issue+ If the 7F;* were to look at their models as
something they were trading with, 'trade unem!loyment and in4ation(, theywould see they ha%e terri&le real time results+ That should tell them something+
I%e &een doing forecasting in the !ri%ate sector since 5==1+ I%e worked on teams
that sold forecasting models and forecasts to lots of maHor industrial and /nancial
com!anies+Mad to relearn what I thought I had learned in grad school in the
5=>8s+ There no+++re!eat no+++com!etition to traditional <lein.ty!e
macroeconometric modeling+ Ne%er has &een+ 'Its not done in s!readsheets, &y
the way, its done in ECiews or something similar(+
This is also true of the !u&lic sector+ Dont let the central &ank thing fool you+
They may look at DSGEs, &ut when it comes to to create the forecast that the
F!en ;arket *ommitttee looks at, the sta goes to 7K0LS, which is more or lessan u!dated traditional model+ And they use addfactors when necessary+ Just like
in 5=R=+ ;eanwhile, *0F and *EALF;0 mainly use traditional models for their
forecasts..and to answer questions from !olicy !rinci!als+ So e%en in the !olicy
world, DSGE is really more of a curiosity than a serious tool+ Fh, and Euro!eans
actually use more structural modeling and know more a&out it than in we do in
the +S+
Incidently, I%e ne%er heard of any&ody using CAKs in a commercial contet+ Too
diOcult to e!lain to decision makers+ ;ay&e some /nance !eo!le do this, &ut I
ha%ent seen it+
I wrote a !a!er a&out this long ago, &ut ne%er could /gure out where to !u&lish
In short, we dont directly use DSGEs to forecast, &ut we do !ay attention to what
they tell us a&out macro and incor!orate that information into our traditional
forecasting tools using add factors and o%errides+ Just &ecause we dont use the
latest tools to !roduce forecasts doesnt mean were stuck in the 68s in ourthinking and su&Hecti%e outlook+
R+
Daniel 'achman5:2B -;
0rian, I agree that current <lein.ty!e models ha%e ado!ted quite a &it from macro
research of the !ast 28 years+ And, yes, you can learn useful things from the
recent literature+
Mowe%er, the question is really tied u! in how the academic world %iews &usiness
and go%ernment forecasting+ )ucas and Sargent !redicted that <lein style models
were on their way out 'in the 5=68s( and that didnt ha!!en+ #ould &e worth
understanding why that forecast didnt !an out, no" 0ut the usual %iew in
academia is that this re4ects a 4aw on the !art of those of us in the industry
rather than a 4aw in the academic research agenda+
Just as an eam!le, right now I am dealing with the fact that it is im!ossi&le to
/nd some&ody trained at school in using <lein.style models+ The only way you
can learn this is %ia a!!renticeshi!+ et these models remain the worlds maintools for using economics in &usiness 'and go%ernment, too(+ I ha%e to untrain
!eo!le &efore I can train them+ #hy isnt there any academic !rogram that takes
!roducing this ty!e of forecast seriously"
Reply
56+
Anonymous=:B5 A;
Ask Ste%e #illiamson to trade with his own models+ ;ay&e we will !air him u!
with Tale&+ Should &e a killer com&ination to raise A;+ And hedging each others
I !ro!ose the following a corollary $CE*; 'or SCAK or whate%er /nance !eo!le
actually use( models are unconditionally &etter for forecasting than DSGE
models$
I still think thats not the same thing as saying that DSGE models dont work+
Fne !ossi&ility is that CAK is as good or &etter than DSGE for forecasting as long
as things are mostly normal 'ie no /nancial crisis and the monetary !olicy rulehasnt changed much(+ I think that descri&es most of the last B1 years+
Ke!ly
5=+
!ohn *. ochrane58:9> A;
Though no fan of DSGE models in their current form, I think you ha%e two &asicfacts wrong+
7irst, &usiness economists who use macro models are not using them !rimarily to
s!eculate on asset markets+ The main task is for indi%idual &usinesses to !lan
in%estment, in%entories, etc+ So they really want !lain %anilla unconditional
forecasts, $whats your &est guess of GD- net quarter"$ -eo!le who want to
make money in asset markets dont use macro models at all+
Second, you are assuming that &etter !olicy conditional forecasts will !roduce
&etter unconditional forecasts+ ou make money o unconditonal forecasts in theend+ 0ut the &itter truth of economic modeling is that this sim!ly is not true+
;odels that make good !olicy !redictions are necessarily clean, trans!arent,
sim!li/ed, and thus dont e!loit all the little correlations in the data to make
&etter unconditional forecasts+ And %ice %ersa, models that do e!loit correlations
in the data fall a!art on making conditional forecasts .. the &ig lesson of the
shifting !hilli!s cur%es of the 5=68s+
ou are wishing this fact away+ Surely good conditional forecasts should hel! us to
construct some sort of melange model that does &oth, you suggest+ 0ut so far
that is a wish not a fact+ The fact is, unstructural use of correlations !roduces&etter unconditional short term forecasts, and sim!le micro founded models
0y the way we are not alone+ -ure $structural$ weather models dont do as well in
making actual forecasts as statistical a!!roaches which com&ine dierent
structural models, sim!le correlations in the data, and forecaster Hudgement+
Ke!ly
Ke!lies
5+
Anonymous55:2> A;
-rofessor . it would &e really interesting to see you e!and on why you arent a
fan of DSGE models in their current form, and if you ha%e any thoughts on howthe !rofession could try to remedy the failings youre thinking of+
I know youre &usy, &ut !erha!s something to consider for a future &log !ost"
1+
hris 'rown55:98 A;
Fn your /rst !oint: there are !lenty of economist ty!es em!loyed at macro hedge
funds and &ond funds who would lo%e to ha%e an edge o%er the market on theim!act of fed !olicy on GD-, in4ation etc+ There is a lot of money at stake here
'the -imco Total Keturn 0ond funds A; [\198&n( and these !eo!le 'to my
knowledge( ignore DGSE so for me Noahs !oint stands+
B+
Won)s Anonymous5:5B -;
Keaders might &e interested to know that *ochrane !re%iously discussed
conditional %s unconditional forecasts at *ato n&ound, touching on Tetlocks$foes %s hedgehogs$: htt!:LLwww+cato.un&ound+orgL1855L86L59LHohn.h.
Id say that itself deser%es a Grum!y Economist !ost+++ ].(
First business economists who use macro models are not using them primarily to
speculate on asset mar!ets.
I know, thats my !oint ;y !oint is that if DSGE models worked, /rms could&edoing this+ The fact that theyre not is %ery informati%e+
"ou are wishing this act away. Surely good conditional orecasts should help us
to construct some sort o melange model that does both you suggest. #ut so ar
that is a wish not a act.
Im not really wishing, Im saying that, F<, su!!ose theres a 7ed !olicy change,
and &uy.side /rms want to know how to &et on the eects+ They &ase their &et on
a model that is aZ&Zc^, where is a weighted com&ination of all a%aila&le
reduced.form models, is a weighted com&ination of all a%aila&le non.DSGEostensi&ly.structural models 'such as sCAKs(, and ^ is a weighted com&ination of
all a%aila&le DSGE models+
#e o&ser%e c8+ oud think that if DSGE models as a class had any %alidity, wed
see c[8, howe%er small+
9+
!e+topiaB:22 -;
Seconding Noah here+ No one is $#ishing the fact away$ 'Id rather like to know,
$why is this fact the case"$(+
$So they really want !lain %anilla unconditional forecasts$
;ay&e Im missing something here, &ut why would &usiness owner assume that a
GD- forecast should &e !olicy in%ariant" If the &est forecast is conditional,
wouldnt e%eryone !refer that forecast, ceteris !ari&us"
$-eo!le who want to make money in asset markets dont use macro models atall+$
And were &ack where we started+
R+
%r&ys B:2R -;
ou can easily enhance returns &y using conditional forecats in entry and eit+ If
#e show that UlooseV monetary !olicy W that is ha%ing an interest rate &elow the
target rate or ha%ing a growth rate of money a&o%e the target growth rate W does
!ositi%ely im!act asset !rices and this corres!ondence is heightened during
!eriods when asset !rices grew quickly and then su&sequently suered a
signi/cant correction+
htt!:LLwww+n&er+orgL!a!ersLw5=9>9
Reply
11+
Anonymous5:98 -;
Id suggest se%eral reasons for this+ 7irstly, /nancial com!anies are run &y !eo!lewho dont ha%e a %ery good intuition for 'macro(economics+ 7or eam!le, a lot of
hedge funds ha%e started assuming that interest rates are not %ery %olatile+
Anyone with a modicum of economics knowledge knows that this is an artefact of
the ^)0: ma'natural rate,83( is always going to &e less %olatile than natural
rate near 83+ Fnce we lea%e the ^)0 these com!anies will &e a &it stued+
DSGEs will only really &e acce!ted if they match these managers intuitions,
which will only ha!!en if they are also &roken and useless+
Secondly, on the other hand, DSGEs tend to ignore a lot of conce!ts like $hotmoney$ that are demonstra&ly %ery signi/cant from a /nancial !ers!ecti%e+ 7or
eam!le a lot of the current &oom in the 0KI*s is !artially dri%en &y highly.liquid
in%estors from Euro!e chasing yield+ The 4ow of hot money into and &ack out of
third.world economies is a maHor cause of their crashiness+
Thirdly, academic DSGEs are normally set u! as toy systems that only handle one
or two economies and a handful of economic indicators+ Ke.tooling them as fully.
4edged world models is !ossi&le &ut non.tri%ial+
7ourthly, and relatedly, with DSGEs I get the im!ression were still in a situationwhere the maths dictates the model+ 7inding closed.form solutions to DSGEs+++
doesnt use them+ Thats &ecause the models were not designed to sell to !ri%ate
industry+ No one is suggesting that those !eo!le should /nd them useful+ #e
design the models for !olicy analysis, and for !ure scienti/c !ursuits . lets try to
understand how the world works+
;ay&e you could write us a !ost a&out how a gold/sh is no good to take on aduck hunt+
B+
Daniel 'achmanB:1> -;
Ah, &ut Treasury and *EA do use ;acro Ad%isors for !olicy analysis+ '*0F uses its
own in.house structural model(+ 0ecause theres no real alternati%e when the
Treasury Secretary or -resident want an answer+ ou might not want that, &ut the
!oint of the original !ost is that...fact...DSGE models arent used much in the$!ractical$ world+ ;y challenge to the academic world is to ask why instead of
sim!ly asumming that we &usinessLgo%ernment ty!es are wrong+
And its not true that &usiness forecasting doesnt need economic theory+ ;uch of
it takes !lace in the medium.term, non/nancial &usiness world, where you need
theory to e!lain 'e+g+( your auto or housing forecast+ Thats theory+
Ste!hen, you are welcome to set u! a com!any to com!ete with the <lein ty!e
models+ <lein did it, after all+ So did )arry ;eyer+ #hy has no&ody from the KE
world e%er come u! with any com!etition in this market"
2+
%r&ys B:95 -;
If the industry doesnt use them, &ut the models actually are &etter then the
currently used in forming conditional forecasts, then its a clear &usiness
o!!ortunity+ The DSGE modeled should form their little hedge fund and !roduce
su!erior returns for a while+ As always, they need to !ut money where their
mouths are+
9+
Noah Smith2:89 -;
So '%m saying it%s clear why private industry doesn%t use them. That%s because
the models were not designed to sell to private industry. $o one is suggesting
that those people should (nd them useul. )e design the models or policy
analysis and or pure scienti(c pursuits - let%s try to understand how the world
The !oint of my !ost is that i DSGE models are indeed useful for !olicy, they
must &e useful for trading as well, e%en if that is not what they were designed for+
See"
And the Ste%e #illiamson quote I !osted says much the same thing Kead whatyou wrote
R+
Daniel 'achman2:55 -;
$Those !eo!le$" I missed that+ Ste!hen, may I suggest that you might not know
that much a&out how the !ri%ate sector use economics if you ha%ent worked in
the industry+ ;y Ho& is not really to su!!ly a &unch of num&ers..it is $to try to
understand how the world works$ so I can e!lain it to decision.makers+ Themodel and forecast are tools to that end+
So if DSGEs really hel!ed me understand &etter $how the world works$ I would
use them+ I dont use them+ #hich is a fact+ And I think thats an im!ortant !oint
to understand+
6+
Stephen Williamson9:5B -;I think were running into a !ro&lem here as to what we mean &y $DSGE+$
5+ Noah seems to think that what I do is somehow re!resentati%e of $DSGE,$ &ut
Im a monetary theorist+ In my academic work, I dont construct large quantitati%e
models that I /t to data+ That said, with reference to the quote from my &log
a&o%e, some of my theoretical work . and things I &log a&out . are suggesti%e of
successful !ortfolio strategies+ Indeed, if I were working on #all St+, I think I could
hel! someone make some money+
1+ #hat ;acro Ad%isers does is Hust /ne+ -eo!le want stories, and that stu can&e used to tell a story, e%en if it means a&solutely nothing+ 0ut there is a market
for stories, a!!arently+
B+ $So if DSGEs really hel!ed me understand &etter $how the world works$ I
would use them+ I dont use them+ #hich is a fact+ And I think thats an im!ortant
!oint to understand+$
So, you dont use them+ 0ut can you tell me what they are"
Meres another o&ser%ation+ Gary Gorton makes, and has made, a lot of money on
#all Street+ 0ut I dont think Gary knows any more than I do 'ask him what he
thinks, if you want(+ Therefore, I think I could get !aid %ery well to work on #all
St+ )arry *hristiano, who I think we could more accurately call a DSGE guy, has a%ery dierent skill set from Gary and me+ Im willing to assert that he could also
make a lot of money on #all St+ if he wanted to+
So, I think !eo!le schooled in modern economics ha%e marketa&le skills+ E%en as
academics, we get !aid ecellent salaries relati%e to what most academics get
'and theyre always com!laining a&out it(+ So, thats the market test+ #hat would
you &e doing with a -hysics -hD, Noah"
=+
Noah Smith=:8R -;
$oah seems to thin! that what ' do is somehow representative o *DSGE* but '%m
a monetary theorist. 'n my academic wor! ' don%t construct large +uantitative
models that ' (t to data.
es, yes, I know+ Technically those models arent dynamic and stochastic+ 0ut I
think the same argument a!!lies to all of the ostensi&ly.)ucas.ro&ust
microfounded modern macro models in use today, and $DSGE$ is Hust a term
!eo!le recogni@e+ Its a &u@@word, sure+
,arry hristiano who ' thin! we could more accurately call a DSGE guy has a
very dierent s!ill set rom Gary and me. '%m willing to assert that he could also
ma!e a lot o money on )all St. i he wanted to.
Im sure he !ro&a&ly could, &eing a smart guy+ 0ut could he do it using the same
kind of model that he makes to try to answer !olicy questions 'i+e+ the kind we
saw him !resent at St+ )ouis(" Thats really the question+
So ' thin! people schooled in modern economics have mar!etable s!ills.
No question a&out that] this !ost wasnt a&out Ho& markets for DSGE.trained
economists+
)hat would you be doing with a /hysics /hD $oah0
Thats a question I%e often asked myself, &ut I guess well ne%er know now
talk a&out the $GE$ too+ Some of the models I work with are not GE in the sense
of com!etiti%e equili&rium, as there can &e strategic com!onents in there+ So,
when you say $ostensi&ly )ucas.ro&ust micro.founded,$ it seems youre talking
a&out the whole /eld of macroeconomics+ So its not clear where your argument
is going+ ou want to argue that all that stu is useless &ecause some !eo!le in
the !ri%ate sector dont want to think a&out it"
52+
Noah SmithB:55 -;
es, all+ ;y critique is of the whole !aradigm+ Theres so damn QmuchQ of this
stu out there that its quite !ro&a&le that a few of the models would in fact &e
%ery useful to industry+ 0ut industry cannon /nd them+ #hy not" 0ecause the
macro !rofession has no way of Hudging models in%alid and throwing them out+ So
whate%er works . if any of the models work . gets lost in the massi%e Hunk !ile,and we kee! Hust adding and adding to the Hunk !ile+
QThatsQ what Im critici@ing+ Its really not e%en a&out the models as much as its
a&out the scienti/c standards of the !rofession that allows all these models to
li%e+ I will write another &ig !ost a&out this fairly soon+++
59+
Stephen Williamson2:1> -;$+++ its a&out the scienti/c standards of the !rofession that allows all these
models to li%e+++$
oure e!ecting something of economic science that it can ne%er deli%er+ This is
Hust the nature of the &east+ I%e learned to lo%e it, &ut o&%iously you ha%e a hard
time with it+ 7irst, in macro, there will ne%er &e one model for e%ery !ro&lem+
#ere trying to understand the world, so the models ha%e to &e sim!le+ ou ha%e
to &reak down this massi%e !ro&lem into little !arts so that we can come to gri!s
with what is going on+ So, there are going to &e a lot of models+ Thats %ery
democratic, dont you think" #e can e%en ha%e %ery dierent models of the same!henomenon, and one might &e as useful as another+ I can like ;o@art, and I can
like Kadiohead too+
The models are all wrong+ ;any !eo!le ha%e em!hasi@ed that !oint 'most
recently )ars Mansen, though I cant remem&er where I read it(+ Thats what
makes economics dierent+ These things are all in%alid, &ut we work with wrong
models &ecause they are sim!le, and . of course . &ecause they are useful+
something a&out how monetary !olicy works] may&e they !ro%ide some insight
into why some !eo!le in the world are so !oor and some are so rich] may&e this
Hust !ro%ides me with another tool . Im going to learn how this thing works, and
Im going to /le that away for later use when Im struggling with some !ro&lem+
So, I ha%e a toolkit . its some array of neoclassical growth modelLsearch andmatchingLmechanism designLinformation economicsLFG modelL)agos.#rightLetc+
etc+ And then I use that for e%erything I do, whether Im writing academic !a!ers,
&logging, writing tet&ooks, gi%ing ad%ice to !eo!le who make !olicy+
0ut there is a class of monster models, which started with 5=R8s )awrence <lein
ty!e 288.equation macroeconometric models+ Those were the models )ucas was
critiquing+ And theres a tendency for more contem!orary models like what came
out of Schmets.#outers and *hristiano.Eichen&aum.E%ans to grow into monsters
that really look much like the )awrence <lein models+ And those monsters are
really quite useless, I think, though the !eo!le who construct those things andwork with them will argue otherwise, of course+
1B+
Stephen Williamson1:52 -;
<r@ys,
Im not sure that were e%en discussing the same thing+ Two questions for you:
5+ #hat eactly is the line of &usiness youre in+
1+ #hy are you reading Noahs &log, anyway"
12+
Stephen Williamson1:11 -;
0( #eirdly, I dont like either ;o@art or Kadiohead+ I know, Im
strange+_M_M_M_M_M_M_M wrong
Didnt say I liked them these characters either+ I assure you that my tastes are
soooo cool that no one likes what I do+
19+
Noah Smith1:29 -;
' you claim to be condemning all o modern macro then that%s essentially
damning all o modern economics to hell as it%s now impossible to say where
Anyway, the cases you quote corres!ond to situations in which the economic
structure is %ery trans!arent, the o!timi@ation does not interact with the
'estimation( noise and the underlying DG- is stationary and %ery sta&le and youclearly control the rules of the game+ Thats why mechanism design a!!lication
'like auctions( work well+ In actual markets, where the fundamentals are not %ery
trans!arent, your a&ility to /lter noise away are %ery limited, econ models will
ty!ically fail+
)et me gi%e you a !erfect eam!le from commodities+ Around 58.59 years ago,
GE was marketing a great !roduct: GE ma!s+ It was great since it !erfectly
ca!tured the fundamentals in the local electricity market in -ennsyl%ania: -J;+
Mow !erfectly" GE ma!s was used &y the system o!erator to actually dis!atch
the system in real time+ So, they knew the underlying su!!ly and demandstructure !retty much with !erfect !recision+ et, they couldnt !redict or e%en
e!lain !ast !rices+ Neither le%els nor %olatilities+ *onsider how strange that is+
Electricity is the sim!lest commodity+ No storage, hence 'almost( no inter.
tem!oral o!timi@ation in the underlying !roduction decision+ 7undamentals as
)ucas *ritique is right &ecause there is no reason to &elie%e that the !arametersof any model you decide to write down are /ed !arameters+ They could change
when !olicy changes+
1+
Som Das$upta2:9R -;
Fk I see the !oint 'and as much as I thought a&out models ha%ing time %arying
!arameters(+
Its interesting and !erha!s the *ritique does not a!!ly uni%ersally 'only to macro
!olicy models !erha!s"(+ The reason I say this is for eam!le we kow that the
0lack.Scholes model with constant %olatility !arameter does not do a great Ho&+
#hich is why im!lied %ols are constantly u!dated and traded+ 0ut that hardly
refutes a model &ased on the no.ar&itrage conditions+
B+
Noah Smith9:51 -;
Some form of the critique a!!lies to most things+
Reply
16+
*an)' 2:B5 -;
Noah, does anyone really $use$ economics to get rich" Mow useful is economics
in actual market a!!lication" Does anyone actually utili@e a macroeconomica!!roach to making money in the markets" I work for a large hedge fund and I
dont know anyone who relies solely on economists and economic analysis for
if the newer is more com!licated without &eing &etter+ Neither ty!e of model
!roduces actiona&le !redictions+ If you dig into 0loom&erg you can /nd data on
how often the a%erage forecast of the economists they sur%ey get right the
direction of this or that macro !arameter, and the track records are all %ery close
to 983+ Some !articular economist &eat that, others do worse, &ut thats
generally %iewed as the natural dis!ersion of the random+
0ut e%ery research team has to take a stand on macro, and clients dont like to
hear that youre Hust kind of guessing+ If old models are as good as any, why toss
them out"
So the more !om!ous !romises made for DSGE indeed fell through, &ig sur!rise+
0ut the /nance industrys $meh$ at DSGE was more a&out con%enience of the
!rocess than the quality of the results+
Ke!ly
Ke!lies
5+
%r&ys 51:B2 A;
Same thing as always+ The &eauty of the street is that if you think you ha%e an
edge that other traders are ignoring, then you trade against them and take their
money+ If you think they are la@y, well, sim!le solution, dont &e la@y yourself and
form the hedge fund already+ ntold riches await+
Reply
B5+
Eric -al)enstein58:2> -;
I was an economists at two &anks way &ack, in 5=>6.>=, and in 5==2.9+
Economists were used for -K, talking a&out econ stats, and sometimes settingthe ta&le for discussions a&out the economy+ The real decisions were made &y
guys winging it a&out macro forecasts+ So, no DSGE, &ut really no macro at all, in
terms of any graduate le%el stu from any thread+
Note they also dont use Stochastic Discount 7actors++++
Nowadays there are lots of !eo!le doing formal macro modeling at &ig &anks,
right" Just not structural, microfounded, DSGE.ty!e stu+
1+
Eric -al)ensteinR:59 -;
The a!e of !ri%ate sector economists was the 5=68s, when e%eryone was
o!timistic a&out large scale macro models+ 0y the time I Hoined a &ank in 5=>6
our &ig &ank had a&out 52 economists, and they would get misty remem&ering
the good old days when there was 98Z !rofessionals de%oted to economics+ After
demonstrating they didnt ha%e any %alue, their ranks slowly dwindled, so when I
was at a &ank in 5==2 we had one economist and me his assistant, and he was
Hust to res!ond to !ress requests are inter!reting macro releases: the &ank liked
getting their name in the !ress+
#hen I was at ;oodys no one inside listened to the com!any economist though
again, he was often quoted in the !ress and inter%iewed, and so remained
%alua&le for that reason+ Needless to say, these -K economists ha%e <eynesian
causal mechanisms &ecause they are more intuiti%e than %ague su!!ly shocks,
regardless of whether they were em!irically &etter, and the whole !oint is to &e
!o!ular with Joe si.!ack+
Fnce, &ack in 5==2, I was sent to )odi Fhio, and was s!eaking to a room full of&ank clients a&out our 'ie, my &osses( macro forecast+ I was &asically a !ro! for
sales !itches &y local &ankers to their clients to sell ser%ices+ I had a lot of
reasons why interest rates would rise, in4ation would rise, etc+ Ff course, if that
was really good info 'it wasnt(, they would ha%e wanted it at our Asset and
)ia&ility *ommittee meeting 'A)*F(, and !erha!s not want to share it+ Fur A)*F
committee couldnt care less what our economist said+ I reali@ed this was dum&,
and got into risk management 'forecasting !ortfolio %olatility(, and later !ortfolio
management, &ecause when done well, they are actually useful+ None of this
uses any macro models+
Reply
B1+
%aleber$51:1R A;
If #Iki!edia is to &e &elie%ed, a DSGE model requires a !roduction function and a
utility function+ That is, you ha%e to ha%e a theory a&out what !eo!le are a&le to
sell and what !eo!le want to &uy+ That seems to &e !ro&lem+ If you can de%elo!
either of those two functions, or e%en &etter, &oth, you can make money withoutin%oking DSGE+ Its sort of like !utting the &ell on the cat+
The other !ro&lem is the E, equili&rium+ The way you make your money in the
/nancial sector is when the equili&rium changes, and when the equili&rium
changes the market is, &y de/nition, not in equili&rium+ All told, Im not sure Hust
what DSGE adds to a /nancial actors usa&le knowledge+
Ke!ly
BB+
Doc ,erlinR:19 A;
This isnt limited to macroeconomics+ -hysics &ased models in s!ace and solar
!hysics also tend to &e far worse than historical &ased 'regressions, looku!.
ta&les, neural nets, etc( models+ Mowe%er, the !hysics &ased models allow us to
learn a&out the su&Hect &y e!loring what is otherwise an o!aque system+
Ke!ly
Ke!lies
5+
Noah Smith58:59 A;
0ut thats &ecause you know the !hysics itself is right+ In econ, if we had goodmicrofoundations . i+e+, a good understanding of how consumers and /rms make
decisions . we could do something similar to what youre talking a&out+
Reply
B2+
Dr. Tufte=:99 A;
Two additions in two comments] /rst addition:
A&out 18 years ago, GewekeQ had a !iece on the scale of innaccuracy caused &y
5( modeling with the )ucas critique &ut ha%ing an unaddressed aggregation
!ro&lem &uilt into your model, %ersus 1( not modeling with the )ucas critique &ut
a%oiding the aggregation !ro&lem 'essentially a <lein model(+ Me concluded that
the scale of the innaccuracies from the two !ro&lems was a&out equal+
This is consistent with the %iew of some in this thread that !ri%ate sector
forecasters may not use DSGE models &ecause that would require an in%estment
of time and eort without any im!ro%ement in outcomes+
Q I think it was Geweke 'in the AEK-?-(+ I couldnt /nd it online on Saturday
The whole article and almost all the comments are a&o%e my !ay grade"
Two questions:
If DSGE model is not usa&le 'in /nancial and in%estment decisions(, why s!end so
much ink on it"
#hy are so many !eo!le cogni@ant of it"
Ke!ly
Ke!lies
5+
%r&ys =:21 -;
Same as with string theory in !hysics: the internal logic takes o%er, the
&ureaucratic logic assures re!lication, r and !eo!le will always /nd a good reason
not to confront their models with data to kee! their Ho&s+
Reply
B>+
%ent Willard58:55 A;
7rom what I%e seen, in%estment decisions are &ased on a !lausi&le narrati%e of
what will ha!!en and why+ If you cant tell that story within 58 minutes and B.!ages, then it wont 4y at the to! le%els+ And that narrati%e had &etter /t the
!olitical %iew, !ersonal e!erience, and rule of thum& of the to! guy+
*om!le models tend to o&scure the narrati%e, so no one is going to stake their
fortune on it+ Nor should they, since too many !rogramming mistakes can &e
made and too many %aria&les arent measured timely or !recisely+ Fnce the
narrati%e has &een chosen, then may&e models will &e used to quantify the
magnitude and timing of the e!ected outcome+ The com!le models are often
used for showmanshi! rather than strategic decisions+
In%estors e!ectations of in4ation for the !ast 2.years were sim!ly &ecause
;0As didnt know the dierence &etween the monetary &ase and the money
su!!ly, and that the latter was created &y commercial &ank lending more than
central &anks+ It was a confusion of terms, they didnt really rely e%en on sim!le
s!readsheet models+
Ke!ly
B=+
daniels5:BR -;
Noah, I think a lot of this is a case of &ig mental adHustment costs+ -ri%ate sector
macro to me still seems dominated &y !eo!le who got most of their training in
the 68s and >8s, at a time when DSGE models were rudimentary+ The state of
the art DSGE modeling with e%erything from /nancial sectors, to search frictions
to all sorts of im!erfect informationLlearning are !ro&a&ly too recent for many!eo!le to &e aware of outside academia and central &anks+ And I think switching
from the simultaneous equations model with all %aria&les o&ser%a&les, to a state
s!ace framework where you aknowledge the im!ortance of latent,uno&ser%ed
state %aria&les and try to measure them is a &ig im!ro%ement in !ers!ecti%e
a&out the macroeconomy+ 0ut things like <alman /ltering are still not !art of the
standard training of most ;asters in economics students I think, so the technical
&aggage required for using DSGE models may &e missing sometimes 'though its
!ro&a&ly there in some quantL/nance a!!lications. e+g for stochastic %olatility
modeling+(+
;acro forecasting is not eactly a &usiness with a great re!utation for accuracyeither in conditional or unconditional forecasting, so while you cant clearly !ro%e
their forecasts and scenario analysis would im!ro%e with using DSGE models, its
hard to argue that they couldnt im!ro%e if they di%ersi/ed their toolkit+
A lot of DSGE models are a&out gi%ing insight, shar!ening intuition &y taking a
more theoretical !ers!ecti%e, which then you can com!lement %ia your
HudgementLadd factors Zstatistical models like CAKs to get a /nal forecast+ Its
not meant to &e the single forecast !roducer, and neither should a large old style
model 'forecast com&ination o%er se%eral models almost always dominates
indi%idual model forecasts(+ And much of the insight can also &e o&tained &y
going o%er all sorts of DSGE modelsL!a!ers, without necessarily using one as akey forecasting tool+
Ftherwise, I sus!ect things like 0ayesian CAKs or dynamic factor models or
sometimes Hust uni%ariat AKI;As will always dominate in !ure forecasting at 5.1
years hori@on+ The more rele%ant question is whether DSGE models can &e useful
in riskLscenario analysis+ ou seem to suggest no, &ecause the microfoundations
are wrong, &ut thats a &it of a straw man+ After all, its not like the ty!ical
consum!tion or in%estment function in a &ig old style models are true either+
Theyre also surely missing many %aria&les 'that must &e the case since theyre
limited to thinking only of o&ser%a&le %aria&les or using %ery rough !roies for
uno&ser%a&le things like !otential out!ut(, using the wrong linear functional form,too hasty in assuming error terms are IID and dont contain richer dynamics, and
Ad%isors and ;oodys Economy+com ha%e com!etition from this direction in
su!!lying tools for stress testing+ 0ecause Im not aware of any&ody su!!lying a
DSGE &ased tool for stress testing+ It does seem like a !otentially lucratu%e
market niche+
1+
daniels2:1B -;
Actually I think ;oodys has started looking at DSGE models as !art of their credit
risk scenarios modeling recently+ 5 !ossi&le ad%antage with DSGE models, is that
with current software such as dynare its quite easy to do for eam!le a Brd order
Taylor a!!roimation of the dynamics, so you can &etter ca!ture the im!ortant
nonlinearities that may matter if youre trying to simulate a more etreme
scenario+ Im not aware of anything equi%alent in the <leinL*owles comission style
models+Im not im!lying &usiness economists are ignorant, though there is a general lack
of con/dence in &usiness economists forecasts 'eem!li/ed for eam!le in the
cha!ter on economic forecasting in Nate Sil%ers recent &ook the $Signa and the
Noise$(. to &e fair forecasting the economy either as a !oint forecast or with
scenariosLwell cali&rated con/dence &ands is going to &e hard whate%er you do+
In academic a!!lications, you ne%er see add factors, &ecause thats not !art of
what you want to do in a Hournal article where the goal is to see how far a model
or theory can go &efore failing without add factors 'I dou&t To&in or Solow or
;odigliani used add factors in any way when writing an academic article using
some ISL); model(+ 0ut in !olicy en%ironments, DSGE models use %ariants of addfactors quite a lot either through the shock !rocesses or %ia e!licit model
com&ination 'treat the DSGE model as a core model, then add etra
dynamicsLadHustment %ia a non.core &lock(, and !resuma&ly youd do the same in
!ri%ate sector macro consulting+ In some cases, the add factorsL/es from the
!ers!ecti%e of one DSGE model can &e Husti/ed as com!ensating for missing
factorsLmechanisms that are !resent in other auiliary DSGE models that are
richer in some dimensions &ut may &e harder to work with+ The &est e!osition I
know of is the recent 0oE e!lanation of their forecasting framework,
This is interesting+ I quite agree a&out the lack of con/dence in &usiness
economists forecast..this is well demonstrated in the literature, as well as &y
Nate Sil%er+ At this !oint, no&ody in the economics world really e!ects much
a&ility in forecasting turning !oints 'educating our customers is another matter(+ This is, of course, where we really need a good tool+ I dont &elie%e DSGEs did
that well in the last downtown+ 0ut how could any econometric model !redict that
the )ehman would fail when it did"
So the question is whether DSGEs !ro%ide another useful tool+ As the original
!ost !oints out, so far this has not !ro%en to &e the case+ 'Again kee! in mind
that while *entral &anks ha%e DSGEs, they also use <lein.ty!e models+ The 0ank
of England, as far as I can tell from what they show on their we& site, ha%ent
re!laced the uarterly ;odel with a DSGE+(
#e had a session on models a cou!le of years ago at AEA 's!onsored &y NA0E, so
it was relegated to an incon%enient location(+ Fne of the 7ed !eo!le at that
session made an interesting !oint: That central &anks 'and es!ecially the 7ed
0oard( ha%e a lot more resources than the !ri%ate sector, so they sim!ly &uy
more of all ty!es of forecasting, without necessarily making strict cost.&ene/t
analysis of what they are &uying+ A nicer way to !ut it might &e that central
&anks are more willing to &uy new, risky technology than the !ri%ate sector+
In any e%ent, I welcome the !ossi&ility that DSGE models could !ro%ide a useful
tool+ I am currently a !urchaser of a traditional model, &ut would welcome a salescall from some&ody who could con%ince me that their DSGE model would &e a
&etter tool+
Any takers" If not, I think the original !ost stands+
Ke!ly
Ke!lies
5+
danielsB:B= -;
I dont get what youre saying a&out a&out the 0oE" The core of their old
forecasting framework was a DSGE model+ The dierence relati%e to a CE*; is
that instead of Hust ha%ing con%ergence to some long run static E, their error
correction terms are res!onding to the de%iation &etween a dynamic equili&rium
model 'a DSGE model, or may&e rather a DGE since it seems the 0E; core is
non stochastic(+ Their new core model is a DSGE model similar to an o!en
economy %ersion of Smets and #outers modelZsome credit constrained
households+ The !ost says this has not !ro%en to &e the case so far %ery frequently in the
!ri%ate sector+ 0ut I dont know that <lein ty!e models were useful in the !ri%ate
sector in the 5=98s either, which didnt mean theyre not useful today+ If I
understand correctly the leaders in using large macro models were a macro
consulting /rm esta&lished &y <lein himself and the 7ed through ;odiglianis ;-S
model+ At the time it must ha%e &een much easier !ro&a&ly to gain clients Hust &y
showing u! with a model when there was nothing com!ara&le+ Nowadays, themarket is still oligo!olistic &ut more crowded, and it would &e harder for someone
like )arry *hristiano to o!en u! a macro consultancy &usiness at the same le%el
as say Fford Economics or IMS: it wouldnt &e enough for him to quit his
uni%ersity Ho&, hed need to /nd do@ens of other like minded economists,
coordinate them, hire secretaries and &usiness de%elo!ment managers+++in short
there are &arriers to entry into the /eld, and like in other 'semi( scienti/c /elds
its not always o&%ious to con%ince non e!ert clients that a new method can do
&etter than something thats &een around for decades and seemed to !ro%ide a
satisfactory though highly im!erfect ser%ice+ 0ut who knows, may&e 0ernanke
and Gertler will get &ored with academia 'not sure what 0ernankes !lans arenow( and theyll team u! to !ro%ide macroeconomic ad%ice &ased on their New
<eynesian DSGEZ7inancial accelerator model" I !ersonally wouldnt mind
listening to 0ernankes macro ad%ice+
0y the way, Nate Sil%er and my comment werent a&out how come the !oint
forecast from &usiness economists didnt !redict )ehman &rothers or other
recessions+ The !oint is that ty!ical !ri%ate sector forecasts seem to &e &adly
cali&rated: their con/denceLuncertainty &ands around the forecast 'e+g in the S-7
density forecasts( are too narrow and underestimate the risks facing the
economy 'and its true this was also an issue for DSGE models estimated on great
moderation sam!les when the /nancial crisis hit( . and why is e%eryone so/ated on the !oint forecast when you know its an 'almost( !ro&a&ility @ero
e%ent" #ith all the uncertainty around, its the con/dence &ands and the
scenarios which should &e more im!ortant+
1+
Daniel 'achman9:88 -;
Daniels:
7irst, I Hust want to thank you for ha%ing such a reasoned discussion+ I a!!reciate
learning from you a&out some as!ects of DSGEs I didnt know+
Fn the 0FE: I got this from looking at their we&site, so may&e I am wrong" 0ut it
does seem to im!ly that they still use their quarterly model, which looked a lot to
me like a small.scale traditional model+ I know for a fact that, at the 7ed 0oard,
the stas main tool remains 7K0LS 'I ha%e a cha!ter on 7orecasting in the
7ederal Go%ernment coming out in a &ook soon'"(, so I%e s!oken to !eo!le there
a&out it(+ Not that DSGEs arent used, &ut when it comes to organi@ing all the
!ieces of the toolkit, the <lein.ty!e model is still at the core+
I dont think you are correct a&out the IF characteristics of the economic
forecasting industry+ There ha%e certainly &een new entries o%er the years...
Dismal ScientistLEconomic+com came out of nowhere to !ick u! a signi/cant
market share in the 5==8s, for eam!le+ And the &ig /rms are always u!dating
their models anyway+ If there was a market ad%antage to &e gained from using a
new methodology, I think it would ha%e &een ado!ted+ I know this &ecause I%e&een there 'and !ushed successfully for the ado!tion of error correction in
estimation, among other things(+ In fact, I heard that one of the large /rms hired
a DSGE modeler at one !oint, &ut they !arted ways after a cou!le of years when
it &ecame e%ident that the /rms in%estment wasnt worthwhile+
It doesnt take $do@ens$ of economists+ Three -h+D+s and may&e one full.time
manager could !ro&a&ly get it started+ Theres !lenty of turno%er among &usiness
economists, so lots of o!!ortunity to /nd !eo!le who are ready for something
new and dierent that they can sell to their users as more u!.to.date than what
others are using+
Agree a&out the cali&ration issue..and es!ecially the !oint forecasts+ 0ut that
leads to the diOculty of getting users to see &eyond the !oint forecast+ Ma%ing
&een there a lot, I can assure you that its non.tri%ial+
ou see, I would like to get &eyond the idea that DSGEs arent acce!ted &ecause
of market failures, or ignorance, or whate%er, and look more closely at how and
why decision.makers use information..and how economics su!!lies it to them+ I
think thats a really worthwhile area of thought that ought to !lay a role in dri%ing
the research agenda+ And I think the sur%i%al of <lein.style models is a reallyinteresting data !oint in that contet+
Reply
21+
Anonymous58:B9 A;
I can s!eak from the model &ased macro hedge trading !ers!ecti%e+
5( ;ost of the &iggest macro funds are fundamental, NFT technical+ And I meanthe ones that are running models+
1( This is a case of ha%ing no faith in the DSGE or 7ed models+ S!eaking from
!ersonal e!erience 'o%er 18 years(, I can say that a lot of the traditional theory
is sus!ect+ Kecent eam!les &eing . no &alance sheet or de&t %iew of the
economy has led to in4ation and growth &eing lower than forecast+ Interest rates
do not work in the real economy as ad%ertised in traditional models+ #hy should I
use them if they do not work" It is clear to us in the real world that the world
does not work like the 7ed descri&es+
B( Dont e%en get me started on currency trading+ The fact that academic models
ha%e such a &ad track record should tell the macro guys something+2( ltimately, the continual denial &y the 7ed that asset !rices do not swing
#hen the 0ank of England in%ited me to gi%e a talk at their worksho! on
macroeconomics, I wasnt sure if they wanted me to !ro%oke 'i+e+ troll( them with
the kind of ske!tical stu I usually write on this &log, or to talk a&out my ownresearch on arti/cial markets and e!ectations+ So I did &oth+ Now, this is a
central &ank e%ent, which means secrecy !re%ails . so I cant tell you what the
reaction was to my talk, or what other !eo!le said in theirs+ 0ut I thought Id
re!roduce !art of my talk in a &log !ost . the !art where I talked a&out DSGE
models+ 'In other words, the !ro%ocati%e !art+(
$DSGE$ is a loose term+ It usually im!lies much more than dynamics, stochastics,
and general equili&rium] colloquially, to &e $DSGE$ your model !ro&a&ly has to
ha%e things like in/nitely far.sighted rational e!ectations, ra!id clearing of
goods markets, certain sim!le ty!es of agent aggregation, etc+ So when I talka&out $DSGE models$, Im loosely referring to ones whose form is &ased on
the 5=>1 <ydland ? -rescott $K0*$ model+
In recent times, of course, K0* models themsel%es ha%e fallen out of fa%or
somewhat in the mainstream &usiness.cycle.modeling community, and ha%e
gone on to coloni@e other /elds like asset !ricing, international /nance, and la&or
econ+ As of 185B, the most $mainstream$ DSGE models of the &usiness cycle are
$New <eynesian$ models+ The most im!ortant of these is the Smets.#outers
model, which has gained a huge amount of attention, es!ecially from central
&anks, for seeming to &e a&le to forecast the macroeconomy &etter than certain!o!ular alternati%e a!!roaches+ If you know only one DSGE model, Smets.
Anyway, my talk asked the question: $#hat can you do with a DSGE model"$
;ost !eo!le who e%aluate the DSGE !aradigm dont focus on this question] they
either trace the historical reasons for the ado!tion of DSGE 'the )ucas *ritique,
etc+(, or they discuss the ways DSGE models might &e im!ro%ed+ Instead, in mytalk, I wanted to take the !ers!ecti%e of an alien econ !rof who showed u! on
Earth in 185B and tried to e%aluate what human macroeconomic theorists were
doing+
A DSGE model is Hust a tool+ Its a gi@mo, like a fork lift or a lithium.ion &attery+
The +S+ and Euro!e ha%e in%ested an enormous amount of intellectual ca!ital .
thousands of !erson.years of our &est and &rightest minds . in creating, testing,
and using these tools+
So what can you do with these tools"
5. -orecast the economy7
Fne thing you might want to do with a &usiness cycle model is to forecast the
&usiness cycle+ DSGE models ha%e im!ro%ed enormously in this regard+ Though
early K0* models were notoriously &ad at forecasting, more recent, com!le
DSGE models ha%e !ro%en much &etter, and are now considered slightly &etter
than %ector autoregressions, and a&out as good as the 7eds own forecasts+
0ut as Kochelle Edge and Kefet Gurkaynak show in their seminal 1858 !a!er,
e%en the &est DSGE models ha%e %ery low forecasting !ower+ *heck out these
These ta&les show the forecasting !erformance for the Smets.#outers model
'which, remem&er, is the $&est in class$( from 5==1 through 188R+ The /rst ta&leis for in4ation forecasts, the second is for growth forecasts+ )ook at the K.squared
%alues+ These num&ers loosely descri&e the amount of the actual macroeconomic
aggregate 'in4ation or growth( that the model was a&le to !redict+ An K.squared
of 5 would mean that the forecasts were !erfect+ oull notice that most of the
num&ers are %ery, %ery low+ The Smets.#outers model was a&le to !redict a &it of
in4ation one quarter out 'though the 7eds internal forecasts were much &etter at
that hori@on(, and not at all after one quarter+ As for growth, the DSGE model had
%ery low forecasting !ower e%en one quarter ahead+
Now, this doesnt necessarily mean that DSGE models are su&.o!timalforecasters+ These things might Hust &e %ery %ery hard to !redict Mumanity may
sim!ly not ha%e any good tools 'yet( for !redicting macroeconomies, Hust like we
arent yet a&le to !redict earthquakes+
0ut theres also some e%idence that we could &e doing &etter than we are+ In this
185B !a!er, Gurkaynak et al+ test the $forecast eOciency$ of DSGE models, and
/nd that their forecasts are not o!timal forecasts+ Also, they /nd that sim!le
uni%ariate AK models are often signi/cantly &etter at forecasting things like
in4ation and GD- growth than the &est a%aila&le DSGE models This is not an
encouraging /nding for the DSGE !aradigm, since AK models are Hust a&out thesim!lest thing you can use+
Also, in this discussion of forecasting, remem&er that the deck has already &een
stacked in fa%or of DSGE models+ #hy" 0ecause of !u&licity &ias and o%er/tting+
If DSGE models dont do well at forecasting, researchers will add features until
they do &etter+ As soon as they do well enough to look good, researchers will
!u&lici@e the success+ This is a !erfectly a!!ro!riate thing to do, of course . its
like im!ro%ing any machine until its good enough to sell+ 0ut it means that the
!u&lici@ed models will ha%e a tendency to o%er/t the data, meaning that their
out.of.sam!le !erformance will usually &e worse than their in.sam!le and!seudo.out.of.sam!le !erformance+
'!date: Cia a commenter, heres a good sur%ey of DSGE models forecasting
a&ility, including how they did in the Great Kecession+ See my new !ost for
more+++(
In other words, DSGE models are !ro&a&ly not %ery good as forecastingtools+++yet+ 0ut theyre a&out as good as anything else we ha%e+ And they ha%e
im!ro%ed considera&ly com!ared to their early incarnations+
2. Gi"e policy ad"ice7
This is what DSGE models are $su!!osed to do$ . in other words, most academics
will tell you that this is the !ur!ose of the models+ Actually, a model can &e
!erfectly good for !olicy ad%ice e%en if its &ad at forecasting+ This is &ecause
forecasts ha%e to deal with lots of dierent eects and noise and stu thats allha!!ening simultaneously, while !olicy ad%ice only requires you to understand
one !henomenon in isolation+
0ut heres the !ro&lem: To get good !olicy ad%ice, you need to know which model
to use, and when+ So how do you choose &etween the %arious DSGE models"
After all, theres a million and one of them out there+ And theyre usually mutually
contradictory] since theyre /tted using many of the same macroeconomic time.
series 'e+g+ +S+ !ost.##1 GD-, em!loyment, and in4ation(, one of them &eing a
good model 'e%en Hust in one s!eci/c situation( means the others must then not
&e good models+
So how do you choose which model to use to gi%e you ad%ice" Fld methods like
$moment matching$, which were used to $%alidate$ the original K0* models, are,
sim!ly !ut, not %ery hel!ful at all+
#hat a&out hy!othesis testing" Again, not %ery hel!ful+ If you make the model
itself the null, then of course youll reHect it, &ecause any model will &e too
sim!li/ed to e!lain e%erything thats going on in the economy+ If you make the
null the hy!othesis that the DSGE model !arameters equal @ero, youll almost
always reHect that null, e%en if the model is grossly miss!eci/ed+
In !rinci!le, I think you should use some kind of goodness.of./t criterion, like an
K.squared, using out.of.sam!le data and adHusted to fa%or !arsimonious models+
At the macro conferences and seminars I%e attended, I ha%ent see !eo!le
saying $)ook at the out.of.sam!le adHusted K.squared of this model #e should
use this one for !olicy$ ;ay&e they do say this, though, and I Hust ha%ent seen
it+ '!date: Mere, some !eo!le, including Smets and #outers, do e%aluate the /t
De/nitely check out this !a!er if youre into macro modeling+(
0ut anyway, theres a few more !ro&lems here+ Fne is the lack of clearly de/nedsco!e conditions] macro theorists rarely work on the diOcult !ro&lem of when to
sto! using one model and start using another 'see net section(+ Another is the
nonlinearity !ro&lem] most DSGE models are lineari@ed, which makes them
easier 'i+e+ !ossi&le( to work with, &ut means that their !olicy recommendations
often dont e%en match the model+
'As an aside, many !eo!le say $F<, we dont know which DSGE model is right, so Hust com&ine a &unch of models, with some weights+$ 7ine+++&ut the weights arent
structural !arameters, so &y doing this you gi%e u! the su!!osed $structural.
ness$ of DSGE models, which is the main reason !eo!le use DSGE models instead
of a s!readsheet in the /rst !lace+(
So to sum u!, DSGE models could oer !olicy ad%ice if you used an a!!ro!riate
model selection criterion, and dealt carefully with a &unch of other thorny issues,
AND ha!!ened to /nd a model that seemed to /t the data decently well under
some clearly de/ned set of o&ser%a&le conditions+ 0ut I dont think we seem to
&e there yet+
6. ,ap from DSGE models to policy ad"ice7
F<, so its really hard to gi%e de/niti%e !olicy ad%ice with DSGE models+ ;ay&e
you could instead use DSGE models as ma!s from !olicymakers assum!tions to
!olicy ad%ice" I+e+, you could say $Mey, !olicymaker, if you &elie%e A and 0 and *,
then here are the im!lications for !olicies and and ^+$ In other words, since
DSGE models are internally consistent, may&e they can hel! tell !olicymakers
what they themselves think can &e done with regards to the macroeconomy+'Another way of saying this is that may&e we can lea%e model selection u! to the
!riors of the !olicymaker+(
Theres Hust one !ro&lem with this+ DSGE models are highly styli@ed, meaning
that its often not possible even to (gure out whether you &uy an assum!tion or
not+
)et me demonstrate this+ )ets take a look at a DSGE model . say, *hristiano,
Eichen&aum, and E%ans '1889(+ This New <eynesian model is %ery similar to the
Smets.#outers model mentioned a&o%e+ Mere is a CEK truncated list of theassum!tions necessary for this model to work:
• -roduction consists of many intermediate goods, !roduced &y mono!olists,
and one single consum!tion good$ that is a *ES com&ination of all the
intermediate goods+
• 7irms who !roduce the consum!tion good make no !ro/ts+
• 7irms rent their ca!ital in a !erfectly com!etiti%e market+
• 7irms hire la&or in a !erfectly com!etiti%e market+
• New /rms cannot enter into, or eit from, markets+
• All ca!ital is owned &y households, and /rms act to maimi@e !ro/ts 'no
agency !ro&lems(+
• 7irms can only change their !rices at random times+ These times are all
inde!endent of each other, and inde!endent of anything a&out the /rm,
and inde!endent of anything in the wider economy+ 'This is $*al%o !ricing$+ The magic entity that allows some /rms to change their !rices is called the
$*al%o 7airy$(+
• The wage demanded &y households is also su&Hect to *al%o !ricing 'i+e+ it
can only &e changed at random times(+
• Mouseholds !urchase /nancial securities whose !ayos de!end on
whether the household is a&le to reo!timi@e its wage decision or not+
0ecause they !urchase these odd /nancial assets, all households ha%e the
same amount of of consum!tion and asset holdings+
• Mouseholds deri%e utility from the change in their consum!tion, not from
its le%el '$ha&it formation$(+ Mouseholds also dont like to work+
• Mouseholds are rational, forward.looking, and utility.maimi@ing+
F<, Ill sto!+ )ike I said, this is a CEK truncated list] the full list is may&e two or
three times this long+
Mow many of these assum!tions do you &elie%e" Im not sure thats e%en !ossi&le
to answer+ 7ormally, most of these are false+ Some are %ery o&%iously false+ The
question is how good an a!!roimation of reality they are+ 0ut how do we know
that either"" Is it a good a!!roimation of reality to say that households !urchase
/nancial securities whose !ayos de!end on whether the household is a&le to
reo!timi@e its wage decision or not" Mow would I e%en know"
In !rinci!le, you could look at the micro e%idence and see which of these
assum!tions looks kinda.sorta like real micro &eha%ior+ Some !eo!le ha%e tried to
do that with a few of the assum!tions of the Smets.#outers model] their results
are not eactly encouraging+ 0ut if you tried to go ask a !olicymaker $#hich of
these things do you &elie%e"$, youd get a &lank stare+
So DSGE models dont make a clear ma! from assum!tions to conclusions+ 0ut
how a&out using them Hust to e!lore the ro&ustness of models to %ariations in
assum!tions" A central &ank 'or the academic macro community( could make a
&unch of DSGE models and com!are their results, Hust to see how dierentmodeling assum!tions aect conclusions+ In fact, thats !ro&a&ly what the
7or forecasting, the common alternati%es are $s!readsheet$ ty!e models '*hris
Sims dismissi%e term( that dont assume structural.ness+ This is the kind of
model used &y the 7ed 'the 7K0LS( and &y some !ri%ate forecasting /rms
like ;acroad%isers+
-olicy ad%ice is the thorniest question, since you need your model to &e
structural+ 7or this, the main alternati%e that has &een !ut forth is called $agent.
&ased modeling$+ I dont know too much a&out this, and the name is weird,
&ecause DSGE models are also agent.&ased+ 0ut &asically what it seems to mean
is to s!ecify a set of microfoundations '&eha%ioral rules for agents(, and then do
a &ig simulation+ The &ig dierence &etween this and DSGE is that with DSGE you
can write down a set of equations that su!!osedly go%ern the macroeconomy,
and with A0; you cant+
So are we wastin$ our time ma)in$ all these DSGE models1 or not7
;y answer is: Im not sure+ So far, we dont seem to ha%e gotten a heck of a lot of
a return from the massi%e amount of intellectual ca!ital that we ha%e in%ested in
making, e!loring, and a!!lying these models+ In !rinci!le, though, theres no
reason why they cant &e useful+ They ha%e 4aws, &ut not any clear $fatal 4aw$+ Theyre not the only game in town, and reali@ation of that fact seems to &e slowly
s!reading, though cultural momentum may mean that the more recently
in%ented alternati%es 'A0;( will take decades to catch u! in !o!ularity, if they
e%er do+
-osted at 2:89 -;
Email This0logThisShare to TwitterShare to 7ace&ookShare to -interest
RB comments:
5+
!E 2:B1 -;
as larry summers said: stick to &agehot, minsky and kindle&erger
$0ut what does that say a&out the use of DSGE models in central &anks] &oth as
a tool for !olicy analysis as well as a tool for forecast generation" #e argue thatin &oth cases the answer is UnothingV+ The /nding that in4ation is not
forecasta&le o%er the Great ;oderation !eriod is consistent with the !redictions
of the DSGE model gi%en the strong monetary !olicy rule estimated for this
!eriod+ S!eci/cally, since under this rule the !olicymaker will alter the interest
rate to counter forecasta&le de%iations of in4ation from the target, the rule will
eliminate forecasta&le mo%ements in in4ation and lea%e only unforecasta&le
shocks to dri%e 4uctuations+ Thus, our /nding of low forecast !erformance is not
necessarily e%idence against the %alidity of the model+$
B+
Nic) Rowe58:26 A;
)ets !ut it another way: if a central &ank is targeting in4ation, the only model
that could forecast in4ation would &e a model the central &ank didnt know
a&out, or was too stu!id to use+
Since in4ation targeting central &anks know a&out DSGE models, and use them,
we can conclude that DSGE models wont &e a&le to forecast in4ation+
2+
RAstudent55:12 A;
According to this logic, CAKs shouldnt work either, yet they do+
9+
Nic) Rowe2:91 -;
KA student: good !oint+ I dont know why CAKS should work+
;ay&e its easier to datamine and o%er/t CAKS than DSGEs"" Is it Hust like those
%iolations of the E;M that disa!!ear as soon as they are !u&lished""
Ff course, if a central &ank is trying to target in4ation at a 1.year hori@on, it may
still &e !ossi&le to forecast in4ation at a less than 1 year hori@on+
chemistry is a se!arate /eld from !hysics+ So that the agents &eha%e according
to mathematical rules does not mean the system has to+
Reply
58+
William AllenB:26 A;
Noah, in your scheme of classi/cation, where would you !lace the stock 4ow
consistent -ost <eynesian models of #ynne Godley and Staord *ri!!s 'New
*am&ridge school(" I /nd these far more 4ei&le than DSGE ty!e models+
Ke!ly
Ke!lies
5+
Noah Smith=:5R A;
#hat does $stock 4ow consistent$ mean" Are DSGE models $stock 4ow
inconsistent$" *an you show me how this is the case" I am curious+++
1+
walt5:9> -;
DSGE models are stock.4ow consistent+ An eam!le of a non stock.4ow
consistent model is ISL);+
B+
#n)nownB:B5 -;
ou are right, what distinguishes so.called !ost.keynesian stock.4ow consistentmodels from DSGE models is not stock.4ow consistency++++ &ut &asically two
general features: 5+ the treatment of micro &eha%ior &y means of aggregati%e
&eha%ioral equations &ased on certain !riors 'which can &e em!irically
estimated( rather than dynamic o!timi@ation .. say, an aggregate consum!tion
function, an aggregate in%estment function, a loan su!!ly function, a to&inian.
ty!e asset demand function, etc]
1+ the adHusting %aria&les 'or model closures( are also dierent .. out!ut adHusts
'instead of !rices( to aggregate demand] asset !rices 'including the interest rate(
adHust to su!!ly and demand in market of stocks of assets, rather than 4ows 'as
in a loana&le funds theory of the interest rate for instance,among other things]
the /nancial system creates money endogenously+++ S7* are also &eing com&ined
with A0; models &y a few !eo!le+
Reply
55+
!onathanB:92 A;
Mi,
weLyou ha%e to &etter understand the main !ur!ose of DSGE modeling:
com!uting, simulating and estimating a microfounded model 'and thus micro
%aria&lesLshocks( with interlinked 'structural( shocks and %aria&les in a stochastic
framework+ This is the main goal and we know that DSGE modeling is unique for
that !ur!ose+
So I ha%e some comments:
5+ 7orecast the economy"
*om!aring to other methods, DSGE forecasting is the youngest 'a&out > or 58
years old( and e%en if it is not $the &est$, it is mainly &etter than what I call $old
methods$+ )et us time
1+ Gi%e !olicy ad%ice's("
It is working %ery well, &ut in a qualitati%e manner 'in a sense that it gi%e us
useful information a&out trends, shocks and dynamic roles of %aria&les( rather
than for a quantitati%e !ur!ose+ It is still hard to !redict the 'eact( %alue of one!oint 'after two or more !eriods(, &ut it is now easy, with a structural
microfounded model, to !redict a trend with res!ect to other structural 'or not(
shocks+
The !ro&lem of the huge amount of models is not s!eci/c to DSGE models+ It is
more related to the research !rocesses and it is a%aila&le for all modeling
methods+++
I am agree with the fact that we need more goodness.of./t statistics &ut we ha%e
already other great tools: mean of estimated shocks, con%ergence statistics,student tests, IK7s, long run and short run %ariance decom!ositions, shared
growth !ath, model &ased detrending, etc+++ the list is may&e as long as your list
in Section B ].(
B+ ;a! from DSGE models to !olicy ad%ice"
ou should ask the following question: what is 'economic( modeling " Assuming
se%eral and acce!ta&le hy!otheses in order to o&tain a light %ersion of the
'economic( fact you want to model+ Again, assuming hy!othesis is not s!eci/c to
DSGE modeling+ *ontrariwise, assuming fatal 4aw's( is common in se%eral other
That &eing said, I think an interesting macro.question right NF# is why is the Seconomy reco%ering 'a!!arently(" I would /nd that su&Hect worthy of macro.
economic analysis Is it the $loose$ 'less tight( monetary !olicy" If so, which
channels worked" Is it /scal e!ansion" Mow, with sequester and stu"
#hat is going on" And if macro models cannot answer that, what good are they"
As to forecasting, nothing can &eat a futures market+ If you want to make money
in the futures market and use a DSGE model to do so, /ne, &ut for the !ur!oses
of a central &ank, any model is strictly second &est+ 7urthermore, it makes no
sense for a central &ank to forcast a targeted %aria&le+ If the forecast is dierentfrom the target, change your actions until they match+
ou might &e interested in this !ost &y Adam Gurri which notes the mismatch
&etween central &ankers academic work and their actual !olicy decisions+
Technical macroeconomics has had %ery little im!act on !olicy, DSGE or not+
Ke!ly
Ke!lies
5+
Noah SmithR:8= -;
't%s not a good sign that '%ve never heard a DSGE model mentioned in policy
debates even amoung highly technically s!illed economists.
Agreed+
&s to orecasting nothing can beat a utures mar!et.
Not necessarily true, since the 7ed has inside info 'note that according to Edge
and Gurkaynak, the 7eds short.term in4ation forecasts are ;*M &etter than any
models, and in fact ha%e an K.squared of 8+2>, which is huge(+
Reply
1B+
danielsR:B2 -;
. !oint forecast accuracy is far from e%erything e%en in forecasting+ 7or decisionmaking it may &e im!ortant to accurately characterise !ro&a&ility of %arious
inter%als or more generally multi%ariate regions for the %aria&les we care a&out+
The &est model for that !ur!ose may tell you theres a lot of un!redicta&le stu
going gi%en the small num&er of o&ser%a&le %aria&les you ha%e+ 0ut it may &e
&etter at quantifying the uncertainty you face than a model that tries to
maimise the accuracy of the !oint forecast at the cost of distorting !ro&a&ility
&ands+
. Fld style large macro models 'what you call s!readsheet models( ha%e tons of
hidden assum!tions hidden in all the @ero coeOcient restrictions they im!ose in
their &eha%ioural equations+ Theyre Hust hidden and harder to unco%er+ And to&oot, its not e%en clear they res!ect the resource constraints and &udget
. ;any of the assum!tions you mention are a!!roimations of the kind you make
in any econometric s!eci/cation 'e+g loglinearity or constant elasticity of
su&stitution, ignoring distri&utional eects , the mono!olistic com!etition
assum!tion that !rices are inde!endent of the num&er of com!etitors in the
market etc+++(+ #e couldnt !ossi&ly e%er &uild a model as a sim!lifying frameworkfor organising our thoughts in such a messy reality if we didnt ha%e to make
these assum!tions+ As we get &etter, we &uild more com!le models with more
realistic assum!tions 'e+g heterogeneous households and /rms, more realistic
wage &argaining etc+++(
. If you end u! seriously working with DSGE models and see many models you
start to get a feel for which model is a!!ro!riate in which situation, and how to
mentally com&ine dierent models each of which may &e missing im!ortant &its
of reality to get a more com!rehensi%e understanding of the economy '&ut isnt
this the same with micro models"(+ ou dont this Hust from the core macro
sequence or ocasionally reading a&out them+. There are ways to classify dierent clusters of models into a smaller num&er of
more general ty!es 'e+g models of /rm le%el credit constraints often ma! into
models of in%estment taesLwedges, other models are &asicaly a&out
taesLwedges on em!loyment, others microfound changes in total factor
!roducti%ity etc+++(+ Some !eo!le at the 0oE should ha%e a lot of interesting stu
to say a&out this+ )ook u! &usiness cycle accounting and an etension &y Kichard
Marrison and co+++also the 0oEs new forecast methodology com&ining a core
DSGE model with a suite of other models 'DSGE or otherwise( is well documented
in a recent working !a!er on their we&site+
. To say DSGE models are not useful in !olicy analysis &egs the question: notuseful in com!arison to what" ;ost of the criticisms that ha%e &een made could
&e made Hust as well a&out the large scale macro models that are su!!osed to &e
useful in !olicy analysis &y !ractioners+ )ucas and Sargent made some good
!oints in the 5=68s+ )ooking at the current s!eci/cation of e+g the ;oodys
macro model or the Fford models most of these !oints ha%e not &een
addressed+
. A signi/cant com!onent of a model &ecoming useful for !olicy analysis is
e!erience in using it+ Theres a reluctance I think on the !art of !eo!le trained in
the old style macroeconomics to switch from simultaneous equations systems
'often estimated &y F)S des!ite the endogeneity &ias( to the more modern states!aceLlatent %aria&les !ers!ecti%e that underlies the DSGE a!!roach+ As more
central &anks and other agencies like the Euro!ean *omission work with DSGE
models, they should &ecome &etter at communicating their insights and using
the etra structure that allows you to see !eo!le, /rms and markets in these
models 'as you said, theyre in that sense similar to agent &ase models( to satisfy
;ate, I lo%e that you are an honest outlet for how economists reason and model
the economy+ 0ut it is a %ery !oor sign a&out the state of the disci!line that you,
a freshly minted star -hD, knows all these limitations to the main modelling
techniques 'since you are o&%iously well trained in them( yet are &asicallyignorant of other a!!roaches that seem to oer quite useful insights+
I am not accusing you of &ias+ ou clearly think things through %ery clearly and
are honest a&out what economists are u! to+ 0ut it seems like economists in
general wont try anything else ece!t tweaking their current tool till it works a
little &etter+
I guess my analogy is that economists are trained in using a hammer to work on
a !ro&lem+ The hammer doesnt seem to work, &ut they wont e%en try a s!anner
or a screwdri%er+ They must modify the hammer until it works a tiny &it &etter+;eanwhile a whole /eld of !eo!le is out there testing other tools and getting
really interesting results+
7rom your writing it seems you are well aware of this+ 7or eam!le you refer to
agent &ased models, &ut dont really know enough to say whether they might &e
useful+ Then a commenter asked a&out #ynne Godley, and again you ha%ent
had much e!osure to this a!!roach+
If you are interested in learning a&out #ynne Godleys a!!roach, his tool&o if
you will, I recommend this &ookhtt!:LLwww+ama@on+comL;onetary.Economics.Integrated.A!!roach.
-roductionLd!L81B8B85>2B
I am also studying an economic -hD and /nd the rather inward looking nature of
the /eld quite distur&ing+ Es!ecially when it comes to macro 'there are many
microeconomists ha!!y to learn from !sychology, sociology etc(+
In any case, when you do read u! on some alternati%e macro a!!roaches I would
really like to read your o!inion 'in my %iew many ha%e similar !ro&lems, some
ha%e dierent !ro&lems(+
Ke!ly
Ke!lies
5+
Noah Smith58:82 -;
Actually, I know a &it a&out agent.&ased models in /nance+++not in
macroeconomics 'remem&er, Im a /nance !rof, I Hust took macro in grad school(+ There are a )FT of dierent things that agent.&ased macro models could &e
e!+ Though note that the equations include some deri%ates of macro aggregates
and at all time !eriods the accounting identities hold+
I thought you might Hust get the &ook from the li&rary to get a feel for why there
eists this grou! who want to reHect DSGE and use other tools+ And of course,may&e outline a &it of a !rosLcons list of DSGE %s monetary models %s A0; &ased
on your reading 'you seem to &e a&le to !ro%ide good clear un&iased %iews on
such things(+
I guess my !ersonal %iew is that there wont &e an a!!roach that forecasts well+
The economy is a com!le system and we know that the !ath it takes is sensiti%e
to initial 'and all time !eriod( conditions+ Add a s!rinkling of !olitics and you dontha%e much to work with
Reply
19+
marcel 58:82 -;
'n recent times o course 5# models themselves have allen out o avor
somewhat in the mainstream business-cycle-modeling community and havegone on to coloni4e other (elds li!e asset pricing international (nance and labor
econ
*an you !oint me to some of the eam!les in la&or" I ha%e an idea I want to
e!lore and it would &e useful for me to ha%e some idea of whats gone &efore+
Thanks
Ke!ly
Ke!lies
5+
Noah Smith58:5R -;
htt!:LL!ress+!rinceton+eduLtitlesL=156+html
#hen Shimer says $neoclassical growth model$ he means $K0*$+
Reply
1R+
daniels 5:88 A;
. I%e looked into many eam!les of old style macro models and A0;s+ The old
simultaneous equations models Hust strike me as documenting a set of
correlations that were found in the data set they used without much structure+
This doesnt matter if you Hust want the &est !oint forecast 'though its &etter
than to Hust use a statistical model like a CAK or e%en uni%ariate AK;Aequations(+ 0ut it does matter if the goal is to answer things like whats the eect
of !olicy or structural shock + As for A0;s I constantly get the im!ression that
the rationality in them is much too &ounded indeed in terms of re4ecting
moti%ations for human actions, and the market !rocesses they model also strike
me as too !rimiti%e+ I a!!reciate the %alue in trying to &uild things from the
&ottom u!, &ut I feel you may end u! with models descri&ing a market !rocess
from hundreds of years ago instead of the so!histicated markets of the 15stcentury+ Theres !ro&a&ly a middle way where A0;s contain more dynamic
&eha%iour considerations and use more so!histicated matching and search
!rotocols+ I can see how such a hy&rid would &e &oth micro founded and contain
more realistic heterogeneity than most DSGE models+ 0ut I dont see anyone
working on this from the A0; !a!ers I%e seen, and a model with millions of
agent ty!es is &ound to oer more com!licated and hard to understand insights
than a DSGE model with 1.2 agent ty!es+
. K0* models ha%e &een a&andoned in the central &ank monetary !olicy
community+ Theyre still used a lot in economic growth,de%elo!ment and taation
analysis+ If it seems they must &e tri%ially wrong &ecause the real world hasnominal !rices that dont mo%e all the time, you should look u! the 0arro critique+
The &asic intuition is that sticky !rices dont necessarily ha%e a &ig
macroeconomic eect if theyre attached to sticky quantities 'and when
quantities mo%e !rices are 4ei&le(+ As for the com!arison of /t: sticky !rice and
wage models ha%e more com!licated equations with du&ious assum!tions a&out
the frequency of wage adHustment+ K0* models ha%e fewer equations,so if you
want &etter forecsting !erformance its easier to s!ecify them with more
com!licated 'realistic"( shock !rocesses to im!ro%e /t+ Its not always clear which
route is more realistic+ 0ut its im!ossi&le to discuss this seriously in &log
with real and com!utational agents] !ath de!endence] !olitical economy]
technological change and economic growth+
In short, the amount of ongoing acti%ity related to A*E is ra!idly growing, and the
a&o%e sites attem!t to gi%e newcomers a &road o%er%iew of this acti%ity+ I ho!e
that the readers of this &log who are not currently familiar with this modeling
a!!roach will &e interested enough to read at least some of the introductory
materials at these sites+
*onstructi%e comments for im!ro%ing the !resentation of materials at these sites
would &e most welcome+
Ke!lyKe!lies
5+
Noah Smith5:98 -;
Thanks, )eigh Ill de/nitely take a look+ I%e &een meaning to read a lot more
a&out A0;+
Reply
B8+
tom$rey 58:8= A;
$In !rinci!le, though, theres no reason why they cant &e useful+$
Mow a&out this !rinci!le: Any model that is successful in !redicting macro
economic &eha%ior, will &e useful only if it allows !olicy makers to change !olicy
decisions &ased on the !rediction+ Such !redicta&le !olicy changes will allowsome market !artici!ants to make money &y !redicting the !olicy changes and
-retend that the national economy consists of a single !erson, the
$re!resentati%e agent$+ This agent owns all the goods, es!ecially all the ca!ital
goods, and does all the work in the economy+ The agent is greedy for material
consum!tion, and la@y+ To consume, which it likes, it must !roduce, which is amatter of indierence, ece!t that to !roduce it must work, which it dislikes+ If it
!roduces more now than it consumes, it can sa%e the dierence as ca!ital goods,
which make its future la&or more !roducti%e+ There are also shocks to
$technology$, i+e+, to how eecti%ely it can use ca!ital to turn la&or into
consum!tion goods] rather &i@arrely, these shocks are &oth negati%e and
!ositi%e, which means that it regularly forgets !roducti%e technologies, and not
&ecause &etter re!lacements ha%e come along+
In addition to &eing greedy and la@y, the agent is is determined to act now so as
to maimi@e not !resent utility, &ut the discounted future stream of utility at all
times 'since it is also immortal(+ 7ortunately, it is incredi&ly foresighted, and
knows the eact distri&ution of future shocks to technology+ 'This distri&ution is
not changed &y anything the agent does] or, if you like, it always acts in such a
way that its e!ectations are eactly ful/lled+( -ossessing unlimited cogniti%e
resources, it is easy for the agent to sol%e the resulting dynamic !rogramming
!ro&lem o!timally+ This will not lead to a smooth !attern of !roduction,
in%estment and consum!tion] if, for instance, there is a &ig negati%e shock to
technology, and shocks are !ersistent, it &ecomes rational to slack o now, and
enHoy leisure] etra work will &e more rewarded later when the agent will ha%e
remem&ered how to do stu+ These 4uctuations are, su!!osedly, the 4uctuations
of the macroeconomy, the &usiness cycle+
I ha%e sketched this sort of model in a deli&erately hostile way, &ecause I think
such things are remarka&ly silly+ 0ut many %ery eminent economists regard them
%ery highly indeed+ ;ostly I think this re4ects &adly on the disci!line of
macroeconomics, &ut it does raise some interesting technical !ro&lems, like:
• Mow do !eo!le e%aluate the /t of these models"
• Do those e%aluation methods make any sense"
• Mow many of the !arameters of these models are actually identi/a&le"
• Mow much data would it take to estimate one of these models to a gi%en
degree of !recision"
• Mow much data would it take to detect that one of these models was
wrong"
• Mow well would these models /t in.sam!le if they were wrong a&out the
structure of the economy"
'ou might well ask $where is the equili&rium, let alone the general equili&rium, ina model with one agent and no trade"$ ou might %ery well ask that+(
DSGE models B empirically irrele"ant and lo$ically incoherent
July >, 185BEditor)ea%e a commentGo to comments
from Lars Syll
Something a&out the way economists construct their models doesnt sit right+
Economic models are often acknowledged to &e unrealistic, and 7riedmanite
assum!tions dont matter style arguments are used to Hustify this a!!roach+ The
result is that internal mechanics arent really closely eamined+ Mowe%er, when it
suits them, economists are !re!ared to hold u! internal mechanics to em!irical
%eri/cation W usually in order to !reser%e key !ro!erties and mathematical
rele%ance+ The result is that models are constructed in such a way that, instead
of trying to e!lain how the economy works, they deli&erately a%oid &oth
diOcult empirical and diOcult logical questions+ This is !articularly noticea&le
with the Dynamic Stochastic General Equili&rium 'DSGE( models that are
commonly em!loyed in macroeconomics
The fact is that DSGE models themsel%es are not Uem!irically rele%antV+ They
assume that agents are o!timising, that markets tend to clear, that the economy
is an equili&rium time !ath+ They use log linearisation, a method which doesnt
e%en !retend to do anything other make the equations easier to sol%e &y forci&ly
eliminating the !ossi&ility of multi!le equili&ria+ Fn to! of this, they
generally dis!lay !oor em!irical corro&oration+ F%erall, the DSGE a!!roach is
structured toward !reser%ing the use of microfoundations, while at the same timein%oking %arious W often unrealistic W !rocesses in order to generate something
resem&ling dynamic &eha%iour+
Economists tacitly acknowledge this, as they will usually say that they use this
ty!e of model to highlight one or two key mechanics, rather than to attem!t to
&uild a com!rehensi%e model of the economy+ Ask an economist if !eo!le really
maimise utility] if the economy is in equili&rium] if markets clear, and they will
likely answer Uno, &ut its a sim!li/cation, designed to highlight !ro&lem V+ et
when questioned a&out some of the more surreal logical consequences of all of
the sim!li/cations made, economists will a!!eal to the real world+ This is not acoherent !ers!ecti%e+
To me this con/rms what I ha%e &een arguing for years now W neoclassical
economic theory is in the story.telling &usiness+
Neoclassical economics has since long gi%en u! on the real world and contents
itself with !ro%ing things a&out thought u! worlds+ Em!irical e%idence only !lays
a minor role in economic theory, where models largely function as a su&stitute forem!irical e%idence+ 0ut Ufacts kickV, as Gunnar ;yrdal used to say+ Mo!efully
hum&led &y the manifest failure of its theoretical !retences, the one.sided,
almost religious, insistence on aiomatic.deducti%ist modeling as the only
scienti/c acti%ity worthy of !ursuing in economics will gi%e way to methodological
!luralism &ased on ontological considerations rather than formalistic tracta&ility+
;odern macroeconomics &uilds on the myth of us knowing the Udata.generating
!rocessV and that we can descri&e the %aria&les of our e%ol%ing economies as
drawn from an urn containing stochastic !ro&a&ility functions with known means
and %ariances+
In the end this is what it all &oils down to+ #e all know that many acti%ities,
relations, !rocesses and e%ents are genuinely uncertain+ The data do not
unequi%ocally single out one decision as the only UrationalV one+ Neither the
economist, nor the deciding indi%idual, can fully !re.s!ecify how !eo!le will
decide when facing uncertainties and am&iguities that are ontological facts of the
way the world works+
Some macroeconomists, howe%er, still want to &e a&le to use their hammer+ So
they decide to !retend that the world looks like a nail, and !retend that
uncertainty can &e reduced to risk+ So they construct their mathematical models
on that assum!tion+
If macroeconomic models W no matter of what ilk W &uild on
microfoundational assumptions of re!resentati%e actors, rational e!ectations,
market clearing and equili&rium, and we !now that real !eo!le and markets
cannot &e e!ected to o&ey these assum!tions, the warrants for su!!osing that
conclusions or hy!othesis of causally rele%ant mechanisms or regularities can &e
&ridged, are o&%iously non.Husti/a&le+ Incom!ati&ility &etween actual &eha%iour
and the &eha%iour in macroeconomic models &uilding on re!resentati%e actors
and rational e!ectations.microfoundations is not a sym!tom of UirrationalityV+ It
rather shows the futility of trying to re!resent real.world target systems with
A gadget is Hust a gadget W and &rilliantly silly DSGE models do not hel! us
working with the fundamental issues of modern economies+
Share this
• Email
• TwitterR
• -rint
• 7ace&ookR
• Keddit
• Stum&le!on
•
*ategories: The Economics -rofession
*omments '9()ea%e a comment
5. paul da"idson
July >, 185B at B:5B !m
Ke!ly
<eynes e!licitly stated this argument on !age 5R of the General Theory when henoted that
Uclassical economists resem&le Euclidean geometers in a non.Euclidean world
who, disco%ering that in e!erience re&uke these lines for not kee!ing straightW
as the only remedy for the unfortunate collisions which are occurring+ et , in
truth , there is no remedy ece!t to throw o%er the aiom of !arallels and to work
out a non.Euclidean geometry+ Something similar is required today in
economicsV+
The unfortunate collisions were unem!loyment] re&uking the lines was re&uking
those who ke!t !rices /ed or sticky
The aioms <eynes o%erthrew, as I ha%e continually argued , were 5 the
neutrality of money aiom, 1 the gross su&stitution aiom &etween liquid assets
and real dura&le ca!ital assets] and B the ergodic aiom+
The /rst two aioms <eynes e!licitly s!eci/es as not in his General Theory and
therefore he has o%erthrown the third aiom follows from his de/nition of
uncertainty and his criticism of ;r+ Tin&ergens method+
<eynes ke!t, and we can kee!, the aioms that '5( !eo!le are self.interested and
try to !rotect their income and wealth] and that '1( /rms try to maimi@e !ro/ts
for surely we can assume that most /rms are not charita&le institutions, andmost !eo!le look out for what they &elie%e is their own self.interest W 'e%en if in a
world of uncertainty they may make decisions which turn out to &e against their
self interest(t
2. paul da"idson
July >, 185B at B:5> !m
Ke!ly
Sorry &ut somehow a few words got dro!!ed from my quote from <eynes+ The
correct !art of the sentence is] U++Euclidean geometers in a non.Euclidean world
who, disco%ering that in e!erience straight lines a!!arently !arallel often meet,
re&uke these lines for not kee!ing straight+V
6. -red Caman
July >, 185B at 58:8= !m
Ke!ly
Tomorrow, or !erha!s the day after, I will argue on the thread UAdam Smith, 7+
^amans K#EK !a!er and the ==3 mo%ementV that <eyness a!!lication of the
Euclidean %s, non Euclidean analogy in economics is critically 4awed and %ery
misleading in regard to what ha!!ens in the real.world economy+ A more realistic
understanding of this analogy in economics will further illuminate the su&Hect of
what the ==3 mo%ement is in !rinci!le+
o da"etaylor5
July =, 185B at =:22 am
Ke!ly
0ut 7red, <eynes was eactly right In Euclidian geometry, if one kee!s on going
in the same direction 'i+e+ doing the same thing( the same things ha!!en+ In the
non.Euclidian geometry of the surface of a s!here, the law of diminishing returns
ha!!ens: the further you get away from your starting !oint the nearer you get to
&eing &ack where you started+ So we%e tamed Natures Hungle and enHoyed its
fruits] &ut &eing stu!id, mankind has not &een grateful for small mercies+ It has
laughed at Schumachers USmall is 0eautifulV criterion of adequacy and continuesto &elie%e Umore is &etterV, des!ite the manifest e%idence that we%e gone o%er
the to! and are reducing a &eautiful and 'when cared for( fruitful world to a few
oases in an industrial desert+
Just &ecause forests are &eing &urnt on the other side of the world where we
cant feel their heat and see their de%astation doesnt mean they are not &urning,
nor make a few oil !alms worth more than the Hungles natural function W
a&sor&ing the suns heat and lifes wastes to cool, calm, cleanse and regenerate
the air and water su!!lies we need to li%e+
USomething a&out the way economists construct their models doesnt sit rightVwhen they dont draw attention to crucial issues like this+
E%en the DSGE models are unsatisfactory &ecause they do not include the wholeof the social system+ They are im!ro%ements on what was a%aila&le in <eynes
time, &ut until some&ody takes Menry Ma@litt suOciently seriously and follows his
UFne )essonV to include the whole she&ang, there will always &e some dou&t+ In
!articular the models &eing used today ha%e still not !ro!erly caught u! with
Adam Smiths claim a&out !roduction de!ending on B factors and yeilding B
returns to these factors 'ground.rent, wages and interest or di%idends(, through
sale of the !roduce+ The signi/cance of natural resources !articularly land is %ital
here &ut Hust a&out e%ery&ody misses it out
I sus!ect the reason for this is that the mono!olists who control the access toland ha%e decided that our uni%ersities should not tell students all a&out what is
going on within the social system and sinces these august &odies !ay the !i!er,
they certainly can call the tune that is !layed through our education system+ The
signi/cance of land within the macro.economy was e!lained &y many of the
classic economists like Da%id Kicardo, &ut the last one to show its im!ortance
was Menry George 5B8 years ago and no&ody cares to read u! his ideas today+
• #ithout ha%ing a com!lete model in hand there is little ho!e that the
e!erts will &e allowed to realise what is the real &raking source on
economic !rogress, and so go%ernments will continue to &e limited in their
attem!ts to / the situation, thank to the interference and sel/shness of