the uniformly most powerful test.” They said, “That™s mathematically

impossible, you can™t do that”we™ve proved that this is the most power-

ful test.” And so statisticians wouldn™t have anything to do with it. Then,

we talked to Abraham Wald, and he initially had the same reaction. But

then he went home and a day later he called and said, “You are right and

I know how to do it and I know what the answer is.”

Taylor: A lot of things followed from that important discovery. And

you had worked out a little numerical example to show that it would

work, at least in some cases?

Friedman: A very simple case, I™ve forgotten what it was. And then

later, one of the jobs we had was to advise the Navy on sampling inspec-

tion. So we got up a whole series of sampling inspection programs

including sequential analysis using those ¬ndings.

One of the other problems, probably the most important one I worked

on, had to do with proximity fuses, which are used when ¬ring an

antiaircraft gun at an incoming bomber or ¬ghter. A proximity fuse is

designed to eliminate the error in timing by being so adjusted that it

would go off when it was near the target. The fuse sends out a radio signal

that would bounce back from the target; if the target was close enough,

the fuse would go off. The radio signal sent out could be adjusted to

different angles and different intensities. What was the optimum design

of the proximity fuse to maximize the chance of hitting the object? A

very interesting problem, and one that we spent a lot of effort on.

Taylor: That sounds like an amazingly complex problem to be work-

ing on. Did you write up papers or reports?

Friedman: Oh, sure. I have those reports somewhere.

Taylor: How did you feel about writing important papers that you

wouldn™t be able to publish, to show to the world?

126 John B. Taylor

Friedman: You can™t conceive of what the situation was at the time.

The war was the most important thing going on and everybody, not me

particularly, but everybody was putting aside almost all other considera-

tions to contribute what they could to help in the war. I don™t think

there was any feeling on the part of any of us that we were concerned

about what would happen to our research. In any event, this was in an

area that was not of much long-term interest for me.

Taylor: What about the methodology of optimization that you used

at the Statistical Research Group. Is that something that you have used

later in economic research, perhaps in your research on monetary policy

rules?

Friedman: I think it comes the other way. The economic view of seeking

an optimum subject to constraints was a way to approach these military

problems, rather than the other way around. But I will say that that was

very interesting because it was so different from anything we had been

exposed to before.

Taylor: Is there anything else that you would like to add?

Friedman: No, I really don™t think there is. The Statistical Research

Group got me involved with a group of people that I wouldn™t otherwise

have been involved with. For example, it was the way I got to know

Jimmy Savage. He and I wrote a number of papers later together.

Taylor: Do you remember how you happened to write the paper with

Savage on utility functions, which gave risk preference at low incomes?

Friedman: I don™t know. I honestly don™t know. Somehow Jimmy

and I must have been talking about it, but I cannot reproduce it. Jimmy

Savage was a real genius, there™s no question that he was a remarkable

character.

Taylor: How did you come to collaborate with him?

Friedman: We got to know one another at the Statistical Research

Group. What happened was that at the time he didn™t know how to write

and I was forced to rewrite some of his papers. He later developed into

an excellent writer. You know, he was almost blind, he could only see

out of one corner of his eye. He was trained as a mathematician, he had

a Ph.D. in mathematics, and then he went on to statistics and really

revolutionized statistics. How we got into the risk paper, I no longer

have the slightest recollection.

Permanent Income Theory

Taylor: Now let™s go on to your research. Let™s start with your research

on the consumption function. I understand that you think that this is

your best purely scienti¬c contribution.

An Interview with Milton Friedman 127

Friedman: I think it is.

Taylor: Could you say a little more about it? Relating to our earlier

discussion, did your early work with data and mathematical statistics help

you develop the idea?

Friedman: Aside from the work I did on the consumer spending sur-

vey in Washington during the 1930s, I also spent several years at the

National Bureau of Economic Research working with Simon Kuznets.

That ended up in the book, Income from Independent Professional Practice.

It served as my Ph.D. dissertation. It was largely statistical and empirical,

dealing with a whole bunch of questionnaires Kuznets had sent out while

he was working at the Department of Commerce. But it also involved

the application of economic theory dealing with the explanation for dif-

ferences of income in different professions. An early venture in the ana-

lysis of human capital.

The book on the consumption function was a combination of ideas

from the professional income study, from the consumers™ spending study,

and the work I was doing on methodology (which ultimately appeared in

the article I wrote on methodology). What I like about the consumption

function book is that it is the best example I know, in my own work, of

the methodological principles that are laid out in my essay on methodo-

logy. You start with a hypothesis. It has implications. You test whether

those implications are correct or not. If the implications are not correct,

you try to adjust your hypothesis and readjust.

In this case I started out with a hypothesis that is similar to that which

underlies the distinction between real and nominal interest rates. How

do people adjust their expectations? How do they decide what fraction of

their income to spend? I developed the hypothesis along these lines. I

put it in a form in which it could be tested and I derived its implications.

I tested those implications and, on the whole, they tended to con¬rm the

hypothesis. I suggested additional tests that should be made to test the

hypothesis. So it was, in this way, methodologically pure.

In addition, it produced a hypothesis that seemed to explain the data.

As you know, the original pressure for the analysis was the apparent incon-

sistency between two bodies of data: long time-series data and cross-

sectional budget data on consumption and income. The question was:

“How could you reconcile those two apparently contradictory bodies of

data?” A lot of hypotheses had been offered to reconcile them. The hypo-

thesis I offered, the permanent income hypothesis, seemed to me a much

more elegant way to rationalize that difference. And it had, as special

cases, almost all of the alternative hypotheses, so it was a consolidation of

a lot of empirical evidence as well as theoretical analysis.

Taylor: It seems to me that your signal extraction characterization of

the problem, as we call it these days, was quite revolutionary at the time.

128 John B. Taylor

Friedman: That really came

out of the work with Kuznets™s

data on incomes from professional

practice. In that earlier work, I

introduced the concepts of per-

manent income and transitory

income in a simpli¬ed form, and

I just carried that right over. In

the professional income data

research, I had three categories:

permanent, quasi permanent

(that™s what I called the inter-

mediate one), and transitory. Later

I got it down to two.

Taylor: Where did you get the

idea to use such statistical decom-

position theories in economics?

Friedman: Just from the fact

that I was simultaneously be-

Figure 6.4 Milton Friedman,

coming an expert in statistical

March 1992.

analysis.

Taylor: I guess it is an example of the bene¬ts of a little crossfertilization.

Your work on the consumption function got characterized sometimes as

kind of an attack on the Keynesian consumption function. Did that