By Philip Pilkington, a writer and research assistant at Kingston University in London. You can follow him on Twitter @pilkingtonphil. Cross posted from Fixing the Economists
JW Mason has an interesting post on the interest rate over at his Slackwire blog. In it he basically tries to resuscitate Keynes’ theory of liquidity preference as that which determines the interest rate on various assets. I think that he does rather a good job given that this is his goal but from the moment I looked into this debate over a year ago I was always bothered by what was going on.
First let us start with the conclusion that Mason comes to when considering the theory of the interest rate that Keynes lays out in the General Theory,
If we take a more realistic view of credit markets, we come to the same conclusion: the yield on a credit instrument (call this the “credit interest rate”) has no relationship to the intertemporal substitution rate of theory (call this the “intertemporal interest rate.”)
Mason gives the example of the mortgage market. He points out, quite rightly, that when a bank loan is made there is no intertemporal substitution on either the side of the bank or on the side of the borrower. What he means by that is that neither the bank nor the borrower must forgo consumption in the present for consumption in the future. So, he asks, what then explains the amount of loans outstanding? He answers as such,
Buying a house makes you less liquid — it means you have less flexibility if you decide you’d like to move elsewhere, or if you need to reduce your housing costs because of unexpected fall in income or rise in other expenses. You also have a higher debt-income ratio, which may make it harder for you to borrow in the future. The loan also makes the bank less liquid — since its asset-capital ratio is now higher, there are more states of the world in which a fall in income would require it to sell assets or issue new liabilities to meet its scheduled commitments, which might be costly or, in a crisis, impossible. So the volume of mortgages rises until the excess of housing service value over debt service costs make taking out a mortgage just worth the incremental illiquidity for the marginal household, and where the excess of mortgage yield over funding costs makes issuing a new mortgage just worth the incremental illiquidity for the marginal bank.
So, again we’re back to Keynes: the interest rate is based on liquidity preference. Now, this looks all neat and tidy doesn’t it? Well, I would argue it isn’t. Let’s wind this back a little bit, shall we?
Okay, so why do economic actors require liquidity? Well, according to Keynes’ theory it is a buffer against uncertainty. Economic actors do not have access to crystal balls — that is, they do not know what is going to happen in the future — so they keep cash on hand in case anything bad and unforeseen happens in the future.
The same is basically true of Mason’s reformulation of Keynes’ liquidity preference theory. The purchase of a house ties the buyer down in a market that is by no means perfectly liquid, while the extension of the loan also places the bank in a less liquid position.
Now, this all sounds good, right? Well, not really. You see the operative word here is ‘uncertainty’. In order for this theory to work we must fully recognise that economic actors operate in an environment of uncertainty. But earlier on in the piece Mason laid out what determines an interest rate as such,
The yield of the bond — the thing that in conventional usage we call the “interest rate” — depends on the risk of the bond, the expected price change of the bond, and the liquidity premium of money compared with the bond. (My Emphasis)
You see those three sections of the sentences I highlight? Well, are they not one and the same thing? After all, isn’t the risk of the bond and the expected future price of the bond inherently tied up with the liquidity premium of money compared with that of a bond?
I do not raise this issue simply to be pedantic. No, I believe it contains within it an important oversight: namely, that these is no such thing as “risk” on the bond in the sense of a given objective probability that people hold inside their heads. If there were then the notion of a liquidity premium on money would be meaningless. Why? Because, again, liquidity is something we desire only in the face of an uncertain future. If we assume that bonds have objective probabilities then there is no uncertain future and there is no need for liquidity preference. Whereas if we assume that the future is uncertain and that liquidity preference is real then there is no such thing as “risk on a bond”.
This makes the whole issue far simpler and also paints the dividing line between the mainstream and the heterodox theory much more clearly: either there is uncertainty, in which case “risk” cannot be measured and so interest rates are subject to “animal spirits”, or there is no uncertainty and interest rates merely reflect intertemporal substitution. There really is no point in talking about liquidity premium and the risk on a bond as the very existence of the former is predicated on the non-existence of the latter.
What’s more, as I’ve argued extensively before, most New Keynesian economists today think that financial markets — i.e. those in which interest rates are set — are subject to animal spirits. This, as I’ve pointed out in the above linked to post, means that any theory of a natural rate of interest is inconsistent with their belief in behavioral finance and economics because the natural rate of interest theory requires that interest rates across the economy are set to line up savings and investment perfectly and thus must incorporate all information about the future perfectly.
Nice argument. Another big problem for neoclassical theory – using risk and uncertainty as if they are synonymous.
Since the 1950s, the Capital Asset Pricing Model has equated price risk with volatility. There is also credit risk, but it is usually set aside as de minimus in discussing Treasuries.
We can’t know how volatile prices will be in the future. But we do know that in the past few decades, 3-year T-notes exhibited an average volatility of about 3 percent, while 10-year T-notes showed a volatility of about 6 percent.
Actual volatilities gyrate around these averages. But longer-maturity Treasuries surely will remain more volatile than short-term ones, while both are highly liquid. Now, how can Keynes et al help me beat the bond market?
“If we assume that bonds have objective probabilities then there is no uncertain future and there is no need for liquidity preference.”
I’m not sure I follow your logic here, Philip (although I agree with your conclusion) A “probability” (at least with a value less than 1) is an measure of uncertainty. If an event was certain, there would be no need for a probability. Also, I’m not sure how often one can determine probabilities “objectively” in economics an finance, since these often are determined by factors that one cannot know with certainty (as opposed to games and various physical processes).
You have to distinguish between risk and uncertainty. Risk is objectively calculable — like a balanced coin being tossed. Uncertainty it not — like where bond yields will be at in 5 years time or whether Chris Christie will win the Republican nominee.
Even if one accepts these rather strange definitions of risk and uncertainty, I don’t think the conclusion about liquidity holds up. Suppose I sell you a bond which you can sell at any future date, with the price governed by a known probability distribution. For example, you have the right to sell it back to me once a month, but the price is randomly generated based on the distribution.
Then the risk is ‘objectively calculable’ since the distribution is known, but there is certainly a liquidity risk since if you suddenly need to sell it, the price you happen to get may be less the expected one (based on the distribution). (If you don’t think that is liquidity risk, then I’m not sure what you think liquidity risk is).
You hedge it. You balance your portfolio using other assets with different risk distributions.
I’m not talking in reality here. But this is how mainstream economics generally deals with portfolio choice.
This isn’t a “strange definition”. It’s the way it’s been used in economics since the 1920s. Frank Knight drew the distinction. Risk can be defined and measured in mathematical terms, while uncertainty can’t be:
Knightian uncertainty is named after University of Chicago economist Frank Knight (1885–1972), who distinguished risk and uncertainty in his work Risk, Uncertainty, and Profit:
Yes I think uncertainty is a key point with Keynes and heterodox economists and with Taleb in wanting to build more redundant and stable systems.
I think this is also the main point of a Job Guarantee Program which is meant to function in a counter cyclical fashion to try and provide a more reliable and steady source of demand.
I think yield expectations also have to be moderated so that we have a sustainable maximum inclusive economy that functions with real interest rates more at the 2-3% level
Interesting play on words. The distinction between “risk” and “uncertainty” is indeed correct, but imho it is not as sharp as I understand that you’re implying. The distinction becomes less “distinct” if you contemplate availability, quality and access to information. Interest rates can be “Rationalized” but how can compound interest be “Justified”?
“… either there is uncertainty, in which case “risk” cannot be measured and so interest rates are subject to “animal spirits”, or there is no uncertainty and interest rates merely reflect intertemporal substitution.”
Why is this true? Even if we distinguish non-quantifiable Knightian/Keynesian uncertainty from other forms of probabilistically quantifiable uncertainty, and insist on the reality of the former, that doesn’t entail that the latter doesn’t exist. If the latter does exist then there are quantifiable risk factors affecting interest rates that are unrelated to intertemporal substitution factors.
Look, it comes down to this: are financial markets characterised by measurable risk or by fundamental uncertainty? They can’t be characterised by both as one logically excludes the other.
That’s just not true. It is possible for both forms of imperfect knowledge to be present at the same time. Keynes, for one, thought so.
Keynes thought that was the case in financial markets? Quote please.
Regarding the actual truth of the matter, can you give me a probability estimate that the S&P will rise by 10% in the next 6 months or can you not? It’s really that simple.
If I short it like I did twice in 2013, the probability it will rise is 100%.
If I go long, the probability it will fall is 100%.
It’s simple! I just need to let yuz know what I’m doing.
Since Keynes described both concepts – fundamental uncertainty and quantifiable risk – in both the Treatise on Probability and the General Theory, the latter in the context of a discussion of the prices of capital assets, then it stands to reason that he thought both were important.
There is no one unified kind of marketable asset covered by the term “financial markets”. If you are pricing a short-term insurance policy whose payoffs are based on stable and reliable actuarial trends, then the element of quantifiable risk will come to the foreground, and the factor of radical uncertainty will be negligible. But if you are pricing some 30-year asset whose payoffs depend on all sorts of political, social, military and technological contingencies that cannot be quantitatively comprehended, then fundamental uncertainty – along with “animal spirits”, irrational and sudden changes in outlook, etc. – is a more prominent factor.
Keynes argued that fundamental, unquantifiable uncertainty was a more important factor than economists and practical investors were inclined recognize, and that we have a variety of ways of rationalizing its presence away and avoiding the topic. But I am unaware of a place where he argues that this is the sole prevailing factor at work in capital markets, and that the whole field of quantitative actuarial science is therefore worthless and inapplicable.
We’re not discussing insurance markets. We’re discussing capital markets.
Post-Keynesians think that capital markets are non-ergodic and subject to uncertainty. Again, the proof is in the pudding: can you give me a probability estimate for the S&P to rise by 10% in the next 6 months or not? It’s a very simple question.
Non-ergodicity doesn’t imply the kind of fundamental uncertainty Keynes and Knight seem to have been talking about. A non-ergodic system is one in which the infinite time average for a system beginning in a specific initial state differs from the ensemble average over the possible states of the system. Usually the claim that some system is non-ergodic is based on a mathematical proof in which both averages are calculated. So it’s not clear to me how ergodic theory in itself sheds light on the distinction between non-quantifiable uncertainty and quantifiable risk. This seems to be a line of thought developed by Davidson, but other post-Keynesians have rejected the application of ergodicity in explicating Keynes’s thinking on uncertainty. See Rosser.
I personally can give no probability estimates regarding the S&P because I am not an investment analyst with the kind of expertise and background knowledge necessary to have informed judgments in that area. But if I were, I suspect I would conclude that the movements of the S&P are not utterly random, and so the probability of a 10% rise can at least be given within some subinterval of [0, 1].
I’m not getting into a debate over what “ergodicity” means.
But do you really think that an objective probability estimate can be given for the S&P example? I mean, you’re not the only one who thinks that. Many investment professionals — usually very poor ones — think that too. But can you justify it. How would you approach the assignment of this probability… I’d be fascinated to know…
I’m not sure I understand exactly what you mean by an “objective” probability estimate. But in any case, I think the collection of large amounts of information about the past gives us a guide to forming rational degrees of conviction about future prospects.
Obviously one piece of evidence would be the number of times the S&P had experienced a 10% increase over a six month period in the past. Then one would seek to identify other conditions, the present or absence of which raises the probability of such increases in the historical record.
If you were forced to make a life or death bet on a conjecture that depended on calculating the likelihoods of various kinds of movements in the S&P, then while you might curse your fate for being thrown into such a situation, I suspect you would also quickly decide there are more rational and less rational ways of approaching the predicament, and that the more rational ways involve some numerical calculations of probabilities.
Returning to my initial point, I am not denying that there are important and highly relevant instances of radical, unquantifiable Knight/Keynes uncertainty affecting investment decisions and other kinds of decision-making under uncertainty. But what I denied is that all financial decision problems fall into this category, and that there are no cases in which considerations of quantifiable risk prevail. As I said, I think both kinds of circumstance can play a role, and the relevance of these different factors varies from one asset to another and one circumstance to another.
Are you serious? And you think that this would give you an objective measure?
Do you not think that analysing the stock market would start… oh, I don’t know… with trying to forecast company earnings? Or are all the people that do this wasting their time? What they should really be doing is looking at how the S&P grew in the past, right?
I’m finding it rather hard to take this argument seriously. I don’t think I’ve ever seen it made before — even by trend analysts.
You didn’t ask about investments based on estimating the future performance of individual companies. You asked about an investment based on estimating the performance of the entire S&P. Attempting to form such an analysis on the basis of a company-by-company analysis of each individual company on the S&P would be hopeless, and the margins of error would multiply and wash out the result. A more sensible approach would be to look at historical patterns in the aggregate.
Okay, Dan. I’m not going to argue with you on this — you can measure earnings in the aggregate, you ever seen a P/E ratio… but nevermind.
Why don’t you try it? Why don’t you try these trend estimates of your’s? Or why not pitch the idea to some hedge fund managers. I can tell you for a fact that most of them aren’t doing it right now. There’s definitely a gap in the market there…
P.s. Regarding Keynes… I think he made a mistake in his chapter on the interest rate. He knew that such risk was incalculable.
“We are merely reminding ourselves that human decisions affecting the future, whether personal or political or economic, cannot depend on strict mathematical expectation, since the basis for making such calculations does not exist.” (General Theory, Chapter 12)
You want a probably estimate that the S&P will rise by 10% in the next 6 moths? Piece of cake: a 0.15 probability of it going up by 10%+/- 2% in the next 6 months.
Ha! Nice one. How did you derive this mysterious probability estimate?
But you know that, surely?
I understand what both of you are saying, but I have to agree with Phil on this one. You are correct that at any given moment, assuming that the market continues normal functioning, risk for a particular portfolio or position can be measured. An investment in Coca-Cola is measurably less risky than one in Dippy’s Sugar Water Co.* for instance. That seems obvious.
However, because A) life is fundamentally uncertain and some unforseeable event could destroy everything at any time; and 2) financial markets are even more uncertain than life in general; and III) financial products have dense and complex inter-relationships that make chaotic “butterfly” effects not only possible but likely, we can say without hesitation that there exists a non-zero probability that, at any given point in the future, all values will have collapsed to zero due to something like the Crash of ’29 or the Crisis of ’08. And not only that, but we don’t even have a way of calculating what that non-zero probability is. .01% chance of total systemic breakdown or 3%? We can’t be sure.
So, while risk can be measured ceteris paribus, we have no way of calculating the chance that the ceteris will actually stay paribus, since we’re dealing with a complex system that, like all complex systems, is prone to chaotic, discontinuous, and unpredictable behavior. See also the Lorenz Attractor and the Abelian Sandpile model. My two centavos.
“… we can say without hesitation that there exists a non-zero probability that, at any given point in the future, all values will have collapsed to zero due to something like the Crash of ’29 or the Crisis of ’08.”
Sure, but that doesn’t mean there is no appropriate range of application for quantitative risk management.
Suppose I offered you a contract to pay you $10,000 if Naked Capitalism is still up and running on January 2, 2014. How much would you be willing to pay for that contract? Shall we say that no amount is rational, because when it comes to whether or not there will be a asteroid impact, nuclear armageddon or sudden global outbreak of bubonic plague “we just don’t know”, and so our expectations of the future are a black void of fundamental uncertainty?
Can you give this contract a probability estimate? If so, how do you derive it? Is it an actual probability estimate or are you just transforming a subjective judgement into an arbitrary number?
If I had some piece of relevant information, I might be able to structure some kind of hedge, which would allow me to make money from my narrow bit of knowledge, despite broad general uncertainty. That is, I could make two or more related bets. If I knew that Dippy’s Sugar Water was gaining on Coke, because of, say, some improvement in Dippy’s production process, but I didn’t have any particular estimate for the direction of the soft drink market in general, I could go long on one, short on the other, and come out ahead, pretty much no matter how the market as a whole moved. My bit of knowledge would profit me, despite my inability to guess the big picture correctly.
If a great many people, with a great many bits of information are participating in well-run financial markets, then the market prices can reflect a lot more information that any individual can manage alone. At least that’s the theory. That no individual, calculating in isolation, can estimate risk is not a refutation; rather, it is an affirmation of the value of financial markets as institutions able to aggregate into risk estimates knowledge, which is otherwise beyond the capacity of individuals to aggregate or calculate.
The problem with the theory is that the practical limit of a financial market’s estimates must fall well short of an ideal, which reaches a limit only in the elimination of all uncertainty. Just as a practical heat engine can reach only some limit of efficiency well short of the conversion of all heat into mechanical work, so a financial market can aggregate only the information available. Even if that is more information than any individual possesses or can handle, it must leave considerable uncertainty. Financial market prices are necessarily point-estimates; only their well-known volatility and skewed or thick-tailed distributions give any hint of a error-bar around that estimate.
In relation to money and monetary policy, it seems to me that the operational inefficiencies and “frictions” of the institutions ought to be the primary concern, but, somehow, neoclassical theory thinks it can abstract away from precisely these concerns with “rational expectations” and still prescribe for the case, with vague gestures regarding ill-identified frictions.
For Keynes, all probabilities were ultimately based on subjective degrees of belief. That doesn’t mean they are all arbitrary. The question is over the degree to which these states of less than perfect confidence are quantitatively tractable and subject to both synchronic and diachronic rationality constraints.
Whether someone will buy the contract described above might depend on some combination of the expected probability they assign to the relevant outcome and their personal degree of risk aversion. The way I understand Keynes thinking on fundamental uncertainty is that there are certain events for which no numerical probabilities for the alternative possible outcomes ( and hence no expected value) can be assigned that are any more rational than any other arbitrary assignment of numerical values.
If you think that some assignments of expected value to a purchase of the contract for $1 are less justified than other assignments, even if no precise measure of expected value is possible, then I would say that is evidence that you do not believe the prospects for the near-term behavior of Naked Capitalism are subject to radical uncertainty.
“If you think that some assignments of expected value to a purchase of the contract for $1 are less justified than other assignments, even if no precise measure of expected value is possible, then I would say that is evidence that you do not believe the prospects for the near-term behavior of Naked Capitalism are subject to radical uncertainty.”
Yes, I do believe that. I just think that you don’t know what I mean by that and are simply redefining the terms to prove your point.
Knight and Keynes, per everything I’ve read, saw risk and uncertainty as distinct with no overlap. Risk is probabilistic and uncertainty, not.
But neither one ever said “quantifiable risk doesn’t exist; there is only radical uncertainty.” So when analyzing the market for some financial asset, it’s not an either/or proposition.
Also, let’s bear in mind that there is an enormous literature on this issue in post-Keynesian economics – and other areas – and there are many competing interpretations as to what Keynes actually meant, and about what kind of model of decision-making under uncertainty best captures his views. Some also argue that Keynes’s views on these matters changed over time.
The first lie in Mason’s The Interest Rate and the Interest Rate is “the”.
Financial markets must deal with both estimable risk and fundamental uncertainty; logical exclusion of opposites doesn’t get them a pass on practical necessity. Financial markets and institutions deal with the irreconcilable by hedging of bets. So, there must be multiple financial instruments, and by implication, an array of interest rates.
Whether there can be gainsaid to be an invisible ideal acting as a strange attractor governing the macroeconomy in the form of a natural rate interest, and representing a collective intertemporal rate of substitution, that it must manifest in our imperfect world not as “the” rate of interest — since there can be no such thing — but as the slope of the yield curve.
Is there a “natural” or normal or policy-desirable yield curve slope? This is how the question should be posed, it seems to me.
If you think in terms of a yield curve, one thing that becomes very, very clear is that a prolonged inversion of the yield curve in a deflation is very undesirable as public policy. And, I think it would be easier to reason out how prolonged flooding of the financial markets with liquidity would tend to promote disinvestment and monopolization.
“I believe it contains within it an important oversight: namely, that these is no such thing as “risk” on the bond in the sense of a given objective probability that people hold inside their heads.”
That may be an argument against Mason, but Keynes made the same point in his “Treatise on Probability”.
Every skillful player of this game can force a draw: therefore, the only winning move is not to play.