Interview with Clive Granger, Nobel laureate for economics since 2003
by Miro Brada
Clive Granger (1934–2009) was, along with philosopher Bertrand Russell and physicist Brian Josephson, another Welshman who won the Nobel Prize - in his case for his methods of analyzing economic time series with common trends. He is also well-known for his statistical model of causality: if the price of morello, in a certain period, affects the estimated price of cherries, then the morello's price cause cherries' prices, in Granger's model.
In the interview, I scrutinized various aspects of economics, including its validity to predict and manage society. To my surprise, Clive refused the '18th century' criterion of the science: that experiment must be repeatable. The uncertainty or triviality of the predictions in economics then serves as an excuse to avoid responsibility, as it was financial crisis 2008, crash of the LTCM 1998, etc. Without objective criteria (like repeated experiments), economics becomes only a 'modern' religion justifying the status-quo in society.
The interview was published in the Czech weekly Týden (Dec. 2004)
Is econometrics applicable to prove God?
- It is said that once a statistician, or perhaps mathematician, visited a King in a far eastern country and was asked to prove the existence of god, he wrote down a very complicated mathematical formula, and said 'there is my proof'. The King not wanting to admit that he did not understand, accepted it.
What's your proof?
- My personal philosophy is that God operates on an entirely different plane to us and therefore there is no way that we can make sensible responses to questions like 'do you believe in God'.
What was your inspiration for your discoveries?
- For cointegration I tried to disprove a remark by a co-worker, but my proof that he was correct suggested implications in various parts of the area in which I had been working.
Does invention in econometrics accord with science, art or rather philosophy?
- I think that a lot of the best discoveries in econometrics involve developments that are both generalisations and also simplifications. These can then be seen to unify new parts of the field, although on occasion a new type of mathematics is required. I think we compare better with scientists than with artists, who are very free-flowing in perspective, or with philosophers who use introspection more than we do.
How much economics depends on politics?
- Politics continually interacts with economics, often in unhelpful ways. Of course the economy can impact an election.
Russia was worse (on average) than West in technological innovations, but better in chess. Does political system determine kind of innovations?
- Under Communism I understand that intellectuals were not allowed to study certain types of data, particularly economic data, so Russia became very strong in the area of Probability Theory (and Mathematics generally) but weak in classical Statistics, with Econometrics almost non-existent, but Operations Research quite strong, such as linear and non-linear programming. I think these were just facts of life in Russia for many years.
Society produces huge amounts of CO2, toxic and radioactive material. We still hear: "there is not enough data to prove it damages our environment"...
...there is never enough data to be 100% certain of something, but it is a question of making decisions under uncertain circumstances. Many people kept smoking despite the increasing evidence that it was dangerous, but equally many others thought that the risk was not worth it and stopped. With Global Warming it will take a very long time to accumulate sufficient evidence for it to be completely convincing to some, nevertheless the evidence is sufficient for many for us to take some preliminary steps, as insurance.
If all statistics confirm something, after it's already evident, it seems trivial...
...if they are all trivial, why is there such a fight to accept them by many economists?
If economics can't repeat experiments, is it a science yet?
- Is meteorology or oceanography a science? The 18th century definition of a science is no longer relevant. The current answer to the question 'is economics a science' is 'who cares?'
Isn't "who cares" a bit arrogant response?
- The only practical reason we would want to be classified as a science is that we may then get bigger research awards. In any other way, the question has few if any relevant features for the practicing economist.
Hume in 18th century wrote: "A precedes B" does not always mean "A causes B". Quantum physics in 20th century has reconfirmed that...
- Hume is correct in saying that 'A proceeds B' is a necessary condition for causality but not a sufficient one. My own definition needs the further condition that A contains information about B that is in no other preceding variable. I believe that time–precedence is a necessary condition for causality and have not seen any 'testable' definition without this requirement.
Could statistics uncover events like WW II, 11.9.2001, Einstein's theory?
- Statistics require a sample of more than one, unless you are a strong Baysian, and so could not make useful statements on the topics mentioned.
Is history derivable from the present? E.g. the default of Argentina (2001), would backwardly determine Black Friday (1929)...
- All time series methods are essentially time non-reversible, relying heavily on the 'arrow of time' and often on 'path-specific' models. Thus, the answer is no.
Could Black Friday repeat itself?
- Black Friday was a stochastic event, it could occur again at any time but has a very small probability.
So the 'stochastic' events such as Black Friday are unpredictable and thus unavoidable?
- Any event is avoidable, or at least its effects can be reduced if it can be successfully forecast. You can ask-- can we forecast an earthquake or a volcanic eruption? It is impossible at present to do that but you can predict the probability of such an event and certain prior events lead to increases in this probability. The same applies to financial crises, such as banking or currency crises. If by 'stochastic' you mean 'somewhat forecastable but not perfectly' then the answer is yes.
Data is static, while reality is dynamic...
- Data is a sample from the distribution, which itself is changing through time. One can only assume that the dynamics is slow enough for the data to allow us to test successfully between interesting alternative models or theories.
Data represents and is - in itself - reality. If data reveals relationship, this relationship instantly becomes the new data (about relationship) - like a snake eating its own tale (recursion)...
- Data itself is far from perfect and econometricians and statisticians have developed 'robust' methods to get around some of these problems. The best "revealed relationships" are mixtures of data analysis and theoretical forms of a specific type. Thus this is only partially data.
Statistics is then maybe too incomplete to be valid and reliable...
- You can never 'prove' anything with statistics but at least one can produce enough evidence to make people change their behavior, e.g. stop smoking!
A blind person don't see the light, although understands its mathematical properties (Husserl's phenomenology). Is reality understandable only through data analysis?
- I think that some theorists are over-optimistic in the extent to which they think they can 'understand' the economy from their model, given that there exist alternative theoretical models and that their model has not been evaluated by the use of data.
Gary Becker, Nobel laureate (1992), claims: higher pricing of cigarettes reduces their consumption. Could data analysis resolve controversies of the death penalty, legalization of drugs / euthanasia?
- Data analysis is certainly applicable if good and appropriate data is available. For example, my study with other economists on the process of deforestation in the Amazon region of Brazil, could only have been conducted with a good panel from that region. I would doubt if Becker's result about the pricing of cigarettes holds up to a comprehensive study! For many of the important topics you mention a suitable data set may not be available.
Could data uncover cyclic causality: A causes B, B causes C, and C causes A?
- In my set up only the past can cause the future, but there can be several causes. Your question has to be framed as A(n) causes B(n+1), B(n) causes C(n+1),and C(n) causes A(n+1), all of which could occur from a data supported model, which is the chain you identify, but with occurances at different times!
Are all methods derivable from one "core" method?
-There is no basic method as there are too many different types of data.
What's the first econometric method?
- Historically, I would guess the chi-square used on a two-by-two table.
Maybe the bootstrap will unite the whole methodology.
- The bootstrap is a useful simulation method to investigate the properties of a given model but is not helpful in suggesting alternative models as the economy evolves.
May different methods applied to the same data, contradict each other? One would conclude: "egg came first", another: "chicken came first".
- Clearly two empirical models could produce contradictory results if one is carefully identified, tested and evaluated and the other is not. It could depend on the set of explanatory variables used, Model A could use one set and model B a quite different set. One obvious case where the problem could occur is with 'structural models' in which the model is built to use constraints implied by a theory, whereas the alternative models has no such constraints. If the theory is wrong the situation of the problem could occur, I think.
Statistical method is like an interface of data and hypothesis. Right?
- I view an interface as the point where two rather different fields that are each well developed find themselves confronting each other on a particular topic, usually an applied one, which has specific restrictions. It is common to find that the two fields have similar concepts but with different names and each has results of interest not known in the other field. Examples could be policy analysis in economics and control theory in engineering or time series econometrics and oceanography.
Is econometrics 'ideologically' connected with IQ tests - psychometrics?
- I doubt it, Adam Smith was talking about empirical relations, I think.
But both use similar abstractions: GDP-intellect, capital-ego, and methods: F-test, chi-square. Why not to merge economics with psychology?
- What drives the methods are the properties of the data rather than the concepts, for example political science has lots of time series and uses our techniques. I agree that there should be more work in common between psychology and economics.
Will software 'develop' economic models on its own?
- If you go far enough ahead all decisions will be made by computers. The program PCGETS already produces adequate dynamic single output/multiple input models and discussions about how such approaches can be generalized are happening.
Does the increasing power of computers improve economic predictions?
- The forecastability of economic variables is largely an inherent property of the variable. The efficient market theory suggests that stock market returns are inherently unforecastable, and subsequent experience suggests that this is correct. Other variables are more forecastable and we have done better as techniques and data have improved, such as the forecasts of electricity demand, particularly daily usage. The most important variables are in this intermediate region, with some hope of improvements in the future but never certainty!
What is your guess about the Euro/Dollar exchange rate?
- Over the next four years the dollar will get relatively weaker, beyond that is too far to forecast at present.
Will Asian economies - China, Japan, Korea - dominate?
- The Asian economies that will eventually dominate are China and India, however not for the next eight years or so.
Are your predictions regarding dollar, China, India, justified by statistical methodology?
- China and India have had the highest growth rates of any country in the last twenty years, they represent a third of the worlds' population and their strategies for further growth do not put them on naturally competing paths. The quantity and quality of data available is not yet sufficient to use sophisticated techniques but a simple plot and extrapolation is enough for general remarks.
