For academic researchers who are readers of this blog (and I know you’re lurking out there), I wanted to call attention to my new paper on Macroeconomic Regimes and Regime Shifts:
Many economic time series exhibit dramatic breaks associated with events such as economic recessions, financial panics, and currency crises. Such changes in regime may arise from tipping points or other nonlinear dynamics and are core to some of the most important questions in macroeconomics. This chapter surveys the literature on regime changes. Section 1 begins with an interpretation of the move of an economy into and out of recession as an example of a change in regime and introduces some of the basic tools for analyzing such phenomena. Section 2 provides a detailed overview of econometric methods that are appropriate for time series that are subject to changes in regime. Section 3 summarizes the ways in which changes in regime can be incorporated into theoretical economic models and briefly reviews applications in a number of areas of macroeconomics.
very interesting. I wonder if you overlooked the work on business cycles of Paul Ormerod in the UK which might fit into this category. Also the work of Acemoglu on “network origins of large downturns” could be a fruitful area.
Jim: Thanks a lot for the heads-up.
Dear Professor Hamilton,
I just heard a talk on something similar at GWU from Michael Owyang, who struck me as a good and honest guy. But one of the issues that struck me during his talk was one that would be relevant for any of these arguments: how do you prove or argue that there are discrete regimes, as opposed to one underlying framework which changes continuously in detail but does not undergo the radical transformation from one regime to another?
Julian
Very interesting piece.
I will leave others to comment on the math.
From my perspective, I would have liked more of a written description and examples of what we mean by ‘regimes’ and changes thereto.
I would argue that a recession is not a regime change. It is merely a business cycle in four parts, and part of the same regime. As we would say a ‘day’ includes a day and a night, so a business cycle should include an expansion and a recession. They’re part of the same regime.
Rather, I think of regime changes as a change in the binding constraint on economic growth. We may consider the inputs to production to be labor (population); capital (money); capital goods (equipment); innovation (technology or other innovation); land; and energy (and other resources, eg, water or weather). At all times, the growth path of the economy will be determined by which constraint is binding, but the binding constraint is not the same all the time. A regime change occurs when the binding constraint changes, along with the associated implications.
So, if we are speaking of intra-cyclical ‘regime’ changes, we’re talking about the balance of supply and demand, which of these is setting prices, and which is growing faster. Within this, there are fight versus flight instincts, with the change from the former to the latter typically involving a collapse of prices (increased price elasticity of demand), and a recession.
Long-term regime changes involve a migration from one binding constraint to another. The associated change in fight to flight attitudes might be longer term (structural) rather than just cyclical, that is, involving either a long expansion or a depression, as the case may be.
That’s an astonishingly Keynesian or even Marxian interest. Many compliments for that!
True “aligned” Economists don’t worry about such piffle because they have a mathematical proof that the economy is always and forever in one regime only of general equilibrium that always results in the best distribution of income (save for government distortion of course) :-). Just the mention above about the existing of “nonlinear dynamics” may result in being anathemized by the AEA, as for true “aligned” economists there is no such thing as “dynamics” and even worse “nonlinear” ones.
«Many economic time series exhibit dramatic breaks associated with events such as economic recessions, financial panics, and currency crises. Such changes in regime may arise from tipping points or other nonlinear dynamics and are core to some of the most important questions in macroeconomics.»
BTW, a point here for my usual pet topic: a huge “regime” change happened around 1994-1995, when a colossal credit bubble was started across the world, and “Many economic time series exhibit dramatic breaks” around that time. But this was not all *directly* associated with negative outcomes, but it was most likely a tipping point. Even if the negatives outcomes of 2001 and 2008 were the eventual consequences of that. So perhaps your attention should also be directed at (seemingly or actually) *positive* regime changes.
This graph in its simplicity is illustrative (and the volume subgraph is terrifying):
http://finance.yahoo.com/q/bc?t=my&s=^GSPC&l=off&z=l&q=l&c=&ql=1&c=^IXIC
This graph on margin debt 1981-2013 also shows a definite regime change in 1994-1995:
http://www.equedia.com/wp-content/uploads/2013/10/spvsnysedebtmargin.jpg
http://www.equedia.com/how-leveraged-is-the-stock-market/
and the private debt-to-GDP ratio for the USA:
http://www.debtdeflation.com/blogs/wp-content/uploads/2008/11/IMG0014_4969953.PNG
http://www.debtdeflation.com/blogs/2008/11/29/can-the-usa-debt-spend-its-way-out/
Thanks for posting this Professor Hamilton,
I’m a professional economist working for government. We recently had a professor come and give us an applied econometrics course in-house. She did a good job , but did not cover the topic of structural breaks very deeply and I have a nagging question with regards to breaks in cointegrating relationships. It would seem to me that given cointegrating regressions are estimated over long period of time, structural breaks for the constant and coefficients will be common. For example, I find breaks on coefficients in a supply for housing regression I recently estimated. If I account for these breaks, the fit of the long-term equations improves dramatically and the error correction term is much smaller. When I estimate my change equation after accounting for the breaks in the long-term equation, the error correction term is still significant, but overall fit of the change equation is not as good.
My intuition tells me that even though one can test for and find structural breaks in cointegration equations, these should be used sparingly. Your thoughts on this subject would be most welcome.
Thanks,
John
John: I would say that if there is a structural break in the proposed cointegrating relation, then it’s not truly a cointegrating relation. Instead, one is claiming that at least at one point in the sample, something happened to cause y1 to deviate permanently from its previous historical relation to y2.
But a separate issue is how to conduct the test for such a structural break. For example, if one estimated the cointegrating relation just by regressing the level of y1 on a constant and the level of y2, the error term is highly serially correlated and the regressor is I(1), features one would want to take into account in doing the test. The simplest way to account for this would be if your model claims for example that z_t = y1_t – y2_t ~ I(0), then look for structural stability of the regression of z_t on a constant and p lags of z_t-j.
Thanks for the reply and tip.
In your opinion, is it the case that few macro time series are truly cointegrated over long periods of time. Take money supply and GDP for instance. There have been a number of well documented breaks in the relationship of money supply and GDP. Does that mean there is no information to be gleaned from the long-term co-movements of the two variables? I was under the impression there is fairly deep literature on identifying structural breaks in regressions of I(1) variables because structural breaks are often encountered when I(1) variables are regressed over long-periods of time…
As an aside, if I do not adjust for breaks and I find the residual of my cointegration regression to be stationary, would that be good enough in confirming that a cointegration relationship exists? No worries if you are too busy to answer these question, perhaps another reader can help.
John
John: In general I would expect that if structural break tests reveal evidence of a break in the cointegrating relation then cointegration tests would usually find that the series are not cointegrated.
I do not believe that money and GDP are cointegrated.
On your philosophical questions, I invite you to take a look at Section 15.4 in my text Time Series Analysis.
if you use a probabilistic model in time, you develop future probabilities of an event based upon the initial conditions of that particular problem. once you identify a point in time to measure your results, you have identified a specific event which no longer carries a probability x, but carries a probability of certainty x=1 since you measured this point. this produces a new mathematical problem going forward with boundary conditions consistent with this new measurement, not the initial measurement at time zero. hence every time you make a measurement, you create a new mathematical problem. you cannot propagate the original problem forward in time and capture different regimes very easily because they may be related to rather low probability events in the original problem. if this is what you are proposing, could you please point out in the chapter where this update occurs? very curious how you do it, since this is done in other fields as well with varying success.
Baffling: In most of the statistical models I discuss, the observer forms the best inference based on the data in hand, which never amounts to 100% certainty about anything. The key equation you ask about is equation (15), which is a probability (strictly greater than zero and strictly less than one) that is a function of the observed data y_t, y_t-1,…,y_1.
You know monetary debasement is a problem when the cost of the material of a unit of currency begins to exceed the value of that unit.
Zinc Prices May Hike Cost of New Coins
JDH I was a bit surprised that you didn’t include the shiny new objects in everybody’s toolbox; viz., the support vector machine (SVM) classification and support vector regression (SVR) models. SVM classification models seem like a natural for identifying when an economy is in or out of a recession. SVM classification models are at the heart of optical character reader machines, facial recognition software, biological & medical classifications, etc. Maybe some clever young grad student will apply it to classifying economic regimes. And SVRs are becoming the “in” thing for businesses as well; e.g., predicting the value of a car after its lease period. In my own neck of the woods SVRs are being used in US and allied countries to predict various failure modes on weapon system platforms under regime changes such as wartime/peacetime optempos, mission pulse/recover status, deployed/garrison, etc. I recently worked such a project with the Canadian Defence Ministry in which we used SVR forecasts to predict umpteen thousands of failures on hundreds of weapon systems. We found that SVR forecasts dominated other linear and non-linear forecasting approaches; e.g., ARIMA models, transfer models, smoothing models, spline models, VAR models, Markov bootstrapping models, STAR models, count models, etc. None of those other techniques were able to adequately handle a “regime change” in the context of military operations nearly as well as SVR models with non-linear kernels.
Obviously, the Fed uses the orthodox economics literature to help it stay ahead of the curve and maintain an appropriate stance.
However, we had a “perfect storm” in the last recession, e.g. from changing demographics, consumption, household debt, “peak oil,” and terminating “too big to fail.”
The US has been in a debt-deflationary “regime” (with very little debt deflation so far as a result of $4 trillion in bank reserves printed by the Fed) of the Schumpeterian depression of the Long Wave Trough, as began in Japan in 1998 and in the US in 1929-31, 1890-93, and 1837-39.
China-Asia and most of the rest of the world is now tipping over and joining Japan, the US, and the EZ.
The Fed cannot raise the reserve rate, as even a modest flattening of the real yield curve spread from 3 months to 5-10 years risks a catastrophic disruption to the term structure and deleveraging of the unprecedented leverage in the global corporate bond and equity markets such that a cascading global, cross-institutional meltdown of the financial system would make 2008-09 seem benign.
The Fed, BOE, ECB, BOJ, and PBoC and the 10-12 largest TBTE primary dealer banks of US gov’t debt have encouraged, enabled, and sustained the largest financial asset bubble in all of world history, primarily as a response to the bursting of the previously unprecedented bubble that they similarly created and for which they are fully responsible (but no institution holds them accountable).
All bubbles burst. The largest bubbles burst spectacularly and with undesirable, even grim, financial, economic, social, and political consequences.
The next business cycle contraction and financial markets bear market risks being “the last recession and bear market” for “capitalism” as we know it (or don’t know it).
Our children and grandchildren, the few that remain, will ask, some with justifiable contempt, what the bloody, bleeping Hades their generational predecessors were thinking and why they permitted (as if they had a choice in the matter) the top 0.01-0.1% to 1% vampire-like parasites to extract in terms of debt-money compounding interest terms in perpetuity the hyper-financialized subsistence life essence and biophysical/thermodynamic basis for existence from the overwhelming majority of the human ape species on our finite, spherical “Spaceship Earth”, and then call it “success”, which the rest of the world was compelled/forced to emulate, irrespective of the costs to the ecosystem, human apes, and non-human species on “Spaceship Earth”.
The top 0.01-0.1% parasitic rentier elite have created, and overwhelmingly disproportionately profited from, a winner-take-all system in which hyper-competition and obscene wealth and income concentration to the top 0.1-1% results in the cost of the top 0.1-1% “having it all” is the bottom 90-99% not having “enough”.
As a species, we can do so much better, but there is no incentive for the top 0.01-0.1% to 1% to care to do so, as they have won the evolutionary winner-take-all contest and have virtually all of the economic, financial, and political power to ensure that the vast majority of their lesser human ape fellows (and non-human species) have not enough to subsist.
Now what . . . ?
Latest from me: http://www.prienga.com/blog/2014/9/23/what-happens-if-ep-costs-continue-to-rise-and-oil-prices-dont
Note the regime change from 2011.