A reader writes (in discussing the Taylor rule):
Like [the] price elasticity of demand, we have an analytical approach that is appealing in theory, but so ill defined as to be useless in practice.
Wow. I haven’t read anything like that since I read Anti-Samuelson. Believe it or not, this was written by a person who purports to do policy analysis.
steven, your comment is filled with the anti ivory tower rhetoric so very common in the “real” world. but it is really not fair to be so critical of “theory” and ivory towers. you want to comment on useless, how about the financial engineers in the “real” world who led us into the financial crisis? theory, even without direct application, usually serves a purpose to elucidate the importance of a particular factor which may not always be directly observable in real world applications.
Baffs –
I am not at all anti-ivory tower. However, the utility of any tool, be that the Taylor Rule or the price elasticity of demand, is to be judged by its ability to underpin action. What should I do, as a policy maker (or investor, as the case may be)? If a tool provides me usable insight, it is a good tool. If it does not, then it is only a curiosity.
If the Haver Analytics Taylor Rule is correct, then we have a narrative for the Obama years, potentially for the Trump years, and causality for the rise of fascism and socialism. It tells us what to do and what to expect and provides insight on actions we might take. If, on the other hand, the Atlanta default model is right, then we have nothing and have to go back to the drawing board for causal relationships.
For academics, well, who cares? But if you’re pulling the trigger on policy, it matters a lot. I have an article pending in which I take Rep Gosar (R, AZ) to task for suggesting a wall would keep out drugs. Do the numbers, and it turns out marijuana smuggling is collapsing on its own and hard drugs are smuggled through official crossing points. You know why, for the latter? Because our analysis suggests drugs are four times more likely to be interdicted by Border Patrol in the open country-side than by Customs at official crossing points. As a result, compact, high value drugs are far more likely to come through official entry points.
That in turn means that market-based visas will be sufficient to close the southern border. Policy comes down to legalizing marijuana at the Federal level (and possibly encourage Mexico to also do so); installing a market-based system for work visas; and providing Border Patrol some rapid response capability to push residual hard drug smuggling towards official crossing points. That’s analysis and that’s strategy, and it will work. But it has huge ramifications. So it had better be right.
So yes, for policy, utility of a given approach matters a lot.
Of course a wall will keep out drugs. At least as well as the Maginot Line kept the Nazis out of France.
The point I’m making non is that you can do all these forecasts and analyses in actual numbers. It turns out that black market businesses — at the US border comprising migrant labor, marijuana and hard drugs — all can be analyzed using plain vanilla business analysis methodologies. It’s not a black box at all.
“However, the utility of any tool, be that the Taylor Rule or the price elasticity of demand, is to be judged by its ability to underpin action.”
this permits you to use tools which are accurate today and blow up with the unknowns of tomorrow, because you are more interested in getting a number to act upon rather than obtaining an understanding of the action. many times theory is not meant to provide a bean counter with an actual number, but to simply improve ones knowledge of an area. steven, you are too open to the idea of a black box. this gets you into trouble in the long run. have you ever obtained a good tool that works every time forever? yet you imply these exist. you may not realize it, but as i said before, you are taking the favorite view of many anti-education folks who despise the ivory tower and criticize theory. but i can’t say that many of those “real world” tools actually exist and work either. while i may not like the guy, the taylor rule has elucidated important relationships that have had an impact on our understanding of the economy.
It depends, Baffs. Some tools work in a wide variety of circumstances, so only in given regimes. For example, I am a pretty good analyst — arguably qualifying as an expert — in supply-constrained markets. In demand-constrained markets, I am largely useless. From July 2015 – July 2017, I had no idea what was coming next. I had zero power to anticipate. Why? Because it was the down part of the cycle, and I lack adequate analytical tools there (as did the rest of the profession, really). So, yes, I am acutely aware that certain tools only work during certain periods. Moreover, for extended periods of time — years! — you can be essentially blind on a market.
OK. So show me the utility.
Here’s the latest STEO’s oil prices, oil supplied, and oil consumed: https://www.princetonpolicy.com/s/STEO-Tab3a-July-2018.xlsx
Now, tell me, if the oil price went up 10%, what do you think would happen to demand? Put another way, is the price elasticity of demand negative, or positive?
Steven Kopits I don’t want to wander too far off the main topic, but you can’t estimate price elasticities of demand with just supply and consumption data. There’s an econometric identification problem, so you need an instrumental variable and some approach like 2SLS or whatever. Of course, you can use time series models to forecast equilibrium points, but that’s not the same thing as estimating the price elasticity of demand. But I’m pretty sure (or at least I hope) I’m not telling you anything you didn’t know already.
Exactly the point. You need so much additional information, that, by the time you’ve worked through it, the price elasticity is essentially meaningless, at least in that form.
Never reason from a price change.
Steven Kopits: The author of this survey seemed to believe that one could estimate demand elasticities, see Table 3 (ungated version here). Take the issue up with him.
Steven Kopits v. James Hamilton on econometrics. Pass the popcorn!
So let’s see what that guy said:
Figure 4 reminds us why it is difficult to be completely convinced by any of these [price elasticity of demand] estimates. Both the supply and demand in any given year t are responding to any of a number of factors besides the current price. Important among these other factors are income (a key determinant of demand) and previous years’ prices. The latter is important for both demand, since it can take many years for the fleet of existing cars to reflect changes in purchasing habits, and supply, since tremendous lead times are required between initial exploration and eventual production. In any given year, both the demand curve and supply curve are shifting as a result of these factors, and one cannot simply look at how price and quantity move together to infer anything about the slope of either curve. The common methodology of including lagged dependent variables in OLS regressions to distinguish between short-run and long-run responses is also problematic.
QED
Steven Kopits: Yes, that’s why we try to do what’s called instrumental variables, and/or include stocks. If we just threw up our hands, we’d get nowhere.
Right, Menzie, but the profession is very far behind the curve. Does the profession know anything about price elasticities during ‘fight’ versus ‘flight’? Why was there a price spike in 2008, but no price spike in 2011-2014, when supply-demand conditions were similar? Why did US oil consumption hold flat during 2005-2008, but give up consumption readily after 2010? That fight/flight toggle is central, and it comes back to questions of deleveraging which in turn is associated with the ZLB and the associated Taylor Rule. And there are questions of supply-constrained v demand-constrained markets and issues of carrying capacity. And if a supply-constrained market, you will have a differentiated response between advanced and emerging economies, because it’s all about reallocation of a fixed level of supply. The US will have a negative price elasticity at the same price for which China has a positive demand elasticity. That’s by definition. And then as Jim points out, you have all sorts of lags and business cycle factors as well as considerations of where you are on the curve (high or low).
So when some economist says, “The price elasticity of demand short run (or long run) is -0.26,” that number is essentially useless for any real world purpose. Because, if you look at the data, in those times when prices were rising, demand was almost always strong — that’s why prices rose! (This is not true of supply shocks, obviously.) And in a recession, demand and prices will be falling at the same time. So not only is the magnitude wrong, the sign is wrong!
So, if you take Table 3 (p 34) of Jim’s survey, each one of these short-term price elasticities of demand is negative. I will bet you in the coming two years, that the elasticity of demand will in fact be typically positive, as it has been generally for the last two years. So what is that table worth in practical terms? How should we use that for forecasting purposes?
Steven Kopits: My guess is that folks like Lutz Kilian have thought about things like this. Suggest you take a look at his work at how a *real* econometrician deals with this sort of stuff (I count myself as a dabbler).
I will bet you in the coming two years, that the elasticity of demand will in fact be typically positive
Huh??? The price elasticity of demand is negative because demand curves are downward sloping. Most people don’t pass on the chance to buy something at a lower price in order to have an opportunity to buy that same item at a higher price. Most people will want to buy more of something at lower prices than they would at higher prices. Just to be clear, the demand curve is NOT a collection of historical equilibrium points (as is the case with something like an IS curve in macro).
For oil, in the next two years, I would guess (dQ/Q) / (dP/P) is going to be a positive number, if you actually put in the numbers from, say, the EIA’s STEO. It has been for the last two years.
Steven Kopits Demand and supply curves are not directly observed. What we observe are equilibrium prices and equilibrium quantities. Demand and supply curves are inferred using equilibrium price/quantity sets and appropriate instrumental variables, which are needed to pin down the curves. If you just regress historical prices against historical quantities using OLS, then you end up with simultaneous equations bias:
https://en.wikipedia.org/wiki/Simultaneous_equations_model
Exactly. What you are saying, Slugs, is that a practical matter, price elasticities of demand are all but useless. That was my point.
What we have available to us in terms of supply, consumption, inventories and prices are essentially all on the STEO table linked above. That’s it.
So tell me how we use the numbers from Table 3, p34, of Jim’s paper, plus the STEO numbers (or if you want to be annual, the BP Statistical Review) to make a forecast for any time period you would care to use. I have literally tried to do that exercise country-by-country for the entire historical record, and I could not make anything of it. Now, I can forecast using other methods, and make very good forecasts in supply-constrained markets. But using price elasticities of demand, it was a complete failure.
By the way, you can make very good long-term forecasts using income elasticity of demand, adjusting for stage of development, anticipated and actual demographics, and type of country. That works. (And that method, by the way, brings China’s steady state consumption to around 50 mbpd, opposed to the 13 mbpd or so today.)
But price elasticity of demand, I have had no success with it for the reasons I have already stated.
Steven Kopits: If one wants to do a forecast, do a VAR or ARIMA. If one wants to do policy analysis (which is what more economists do), then one wants to examine how prices and quantities respond to an exogenous change, holding other factors constant (what is known as ceteris paribus). Then at a minimum you want a reduced form model, but preferably a structural model, where you have some idea of the structural parameters.
Steven Kopits I don’t know diddly squat about oil markets, but it sounds to me like you’re referring to “price elasticities of demand” when what you really mean is something along the line of an impulse/response forecast. Econometrically that’s a different kettle of fish. Menzie suggested something like a VAR or ARIMA model. As I said, I don’t know diddly about oil markets, but intuitively I would think that the supply and demand responses may not be symmetric and the differences may not be linear responses to shocks, so instead of a VAR or ARIMA model (which are linear), you might have more success with a non-linear model; e.g., a threshold autoregression (TAR) approach. It’s easy enough to test for non-linearity; the tough part is in trying to identify the type of non-linearity. In any event, my main point is that forecasting models like a basic VAR are atheoretical by design, so they aren’t intended to answer questions about structural parameters like the price elasticity of demand. To determine a structural parameter you need an structural econometric model.
So, when we’re talking price elasticities of demand and noted macroeconomist refers to himself as a ‘dabbler’, then we’re in trouble.
I agree the two best academic guys in the field are Jim and Lutz, and I have read most of their work. The difference between them and me is that I am in the markets every day. I know the historical record as well as either of them, and I also know what has been going on for the last four days. Jim and Lutz write papers. I am a huge fan of both of them. But my success or failure hinges on the quality of my calls, so I need tools that help me see the road ahead.
In my view, the price elasticity of demand is true and correct as a theoretical construct. But in oil markets, given the data available, I can’t use it effectively to make a forecast.
Steven Kopits: Let’s say for a second there are no demand curves in oil markets. There are a heckuva a lot of other markets in the world.
By “dabbler”, I mean a dabbler in econometrics. I *apply* econometric methods, I don’t develop new ones.
Right, but you get my point.
Models today are not developed that way. Rather, they are the result of big data and machine learning, which again uses an essentially evolutionary approach. It gets a bunch of data and then goes iteratively through lots of different models to obtain a good result. But as I understand it, a lot of these models are essentially black boxes. Their owners don’t necessarily know how they work or how to fix them if they don’t.
In many cases, these black boxes can out-perform fund managers and analysts. The problem is, of course, that if they don’t work, their respective managers are at a bit of an impasse. How long can a black box hold a losing position until the responsible fund manager bails? Courage is a function of knowledge, and if you don’t know how good your model is, how much courage can you have in defending it? And for how long?
if you don’t know how good your model is, how much courage can you have in defending it? And for how long?
Okay, here I can feel your pain. A few years ago a Canadian Defence Ministry analyst and I were tasked to look at some black box machine learning approaches in support of some NATO agreement. In particular, we looked at the predictive accuracy of a support vector regression (SVR) approach, which belongs in the family of support vector machine (SVM) algorithms. I can follow the math, but I can’t honestly claim to understand the basic intuition in the SVR. Still, in several courses-for-horses out-of-sample competitions across different NATO countries, the training cross-validated SVR model with a radial kernel won hands down and by large margins and across all kinds of regimes. But to this day the results of that study are sitting on a shelf (actually inside a CD-ROM) collecting dust because no one in leadership trusts a model that can’t be explained in an intuitive way. But then again, those SVM approaches are at the heart of today’s facial recognition algorithms.
https://www.mathworks.com/help/stats/understanding-support-vector-machine-regression.html
Didn’t “Princeton” Steve tout some Haver version of the Taylor rule? As if Haver’s version is the only correct version I guess.
You did ask Princeton Steve to read the vast literature on the Taylor rule. I guess he hasn’t given his latest incredibly bizarre comment.
Some have wondered if Princeton Steve is advertising for us to read his blog. No thank you as his blog is surely another waste of time.
Well, thank you, pgl.
You can find my blog here: https://www.princetonpolicy.com/ppa-blog/
If you are interested in illegal immigration as a problem to be solved rather than just ideological talking points, I would consider my blog a very good source of information.
I’ve read your writing and I have long been convinced to skip your blog based on your writing. I think I’ve made that clear. Assuming of course you passed 3rd grade English.
Here is the dated article I was talking about It is 1998!!
https://www.frbsf.org/economic-research/files/3-16.pdf John Taylor was not amused
Note the two articles they cite:
Taylor, John B. 1993. “Discretion Versus Policy Rules in Practice.” Carnegie-Rochester Conference Series on Public Policy 39, pp.
195–214.
1997. “An Historical Analysis of Monetary Policy Rules.” Unpublished paper, Stanford University (December).
Two papers Princeton Steve refuses to read.
Of course even Taylor would note Federal Reserve policy in the 1970’s was an enigma.
I skim read portions of the incredibly long critique of Samuelson and just was amazed. It was sort of like reading some Bernie Sanders cheerleader going off on people like Paul Krugman. I had to stop because reading that gibberish is hazardous to one’s brain cells. Although it is a wee bit more enlightening than the rants from our Usual Suspects.
The University of Michigan lent its good name to this babble?
PoMo as in post-modern? Not the aboriginal people from northern California? My wish: before I die, I actually manage to understand what post-modernism is.
If Steven Kopits has trouble with the Taylor rule ‘theory’ and demand elasticity measures/estimates, I wonder what he thinks of purchasing power parity (PPP) exchange rate accounting/theory or for that matter, modern biology and ecology.
My own personal beef is the way lay people, journalists AND some academic economists use consumption and demand as if the two were synonyms.
My wish: before I die, I actually manage to understand what post-modernism is.
Anything French.
PPP is kind of infuriating but largely unavoidable. So, developing countries will often see rapid appreciation in their exchange rates, but far less in terms of on-the-ground changes. It’s a pick-your-poison kind of thing. I think you can more or less use the GDP measure you want depending on the circumstances. Not perfect, but in my experience, workable.
As for evolution, economics and ecology:
From my perspective, an economy is a monetized ecology, and therefore the key drivers of evolution — which are closely related to a given ecology — should also be key drivers of economics. Of these, some important ones are:
– fight v flight, greed and fear, trial-and-error. Recurring theme, not well understood in economics, probably a Nobel prize in there somewhere. So depression = flight, and behavior in flight (eg, price elasticities of demand) will be different than during fight mode. Like in poker: Do you call and raise, or fold? It depends on the circumstances and on your mood.
– increasing returns to scale – this is the key to principal-agent theory, the rise of language and groups, that is, conservative economics in which we use the group as the unit of analysis. Increasing returns to scale also create politics and dictatorship
– declining marginal utility of wealth and income – the key to a civilized society and the decline of dictatorship
– incrementalism – there always has to be some plausible, small step between one level of development and other. The system as a whole can be complex; the evolution of any single component of the system has to be simple to be plausible from a probabilistic point of view
That some of it. I am the kind of guy who will look at bluejays vs cardinals and try to figure out the respective market niches and associated economics. Or, for example, you may watch Narcos as crime drama. I watch it as microeconomics.
Dani Rodrik’s book Economist Rules is a must read. But I’d bet the ranch Princeton Steve has not read this either. Until he does – I’m skipping anything he bothers to write anywhere.
What do you want to know? Models are a good thing. Sure. I have lots and lots of models.
You read Rodrik and co. and you read lots about liberalism and neoliberalism. And that’s great. But what economists ultimately do is optimize a given objective function. That’s why all the math.
But economists do not own the objective function. Politicians do. Economists work for politicians, not the other way around.
So here’s my question. Half the advanced countries’ population describes themselves as conservative. What is the objective function of a conservative? When I see Dani Rodrik write something on that, then he’ll have my full attention.
Dear Menzie,
I don’t want to get into the strain of analyzing instrumental variables online, (it is apparently necessary to try to understand “.do” files in Stata to do them elegantly), and think the comment you are responding to about the Taylor Rule is idiotic. But sometimes, I can talk about English. Aren’t you missing a “since”, as in “since reading Anti-Samuelson”.
Hoping to get one thing right,
J.
Julian Silk: OOps, yes you are absolutely right. Thanks! Fixed now.
P.S. There should be a question mark at the end. Sigh. J.
A couple of more thoughts on price elasticity of demand, as it occurs to me that I have spent some considerable time on this topic:
Carrying capacity is defined as the region where (dQ/Q) / (dP/P) <= -1. The -1 limit defines the maximum revenues which the industry as a whole can obtain from customers, and in 'flight' mode constitutes about 5.1% of world dollar GDP. In 'fight' mode, this number can be about 50% higher for a short period of time, or approximately 7.5% of world GDP. In dollar terms, it's about $135 and $200 / barrel respectively.
The most important determinant of the price elasticity is the position on the curve. This can be seen by visual inspection of a stylized downward-sloping demand curve. This is a quibble I would have with Jim: Elasticities need to be stated at a given P, not for all P — and determining this is no mean feat. I do, in fact, have such elasticities, some informal and some more formalized, for various countries and various sectors. For example, at $55 / barrel, US per capita (but not total) driving is already peaking. Global oil demand growth at, say, $100 / barrel, should not much exceed 1 mbpd. Elasticities for substitutable goods are much higher than for monopoly goods. Thus, for example, gasoline usage fell by about 4% in the last round of high prices; heating oil usage fell by 50%.
Having said that, price elasticities of demand, as a stand alone tool, do not at all give the results textbooks would suggest. As I note above, demand growth is pro-cyclical, and this prices tend to rise in periods of strong demand. Similarly, both prices and volumes will tend to collapse during a recession. In both cases, the sign of the price elasticity will be positive, not negative as the textbook would have it. Thus, price elasticity of demand, taken as a stand-alone indicator based on available data (prices, consumption, supply and inventory), cannot be used reliably for forecasting price – volume relationships, either in the short or long run.