In Which: Dr. John explains why it took a year to run his first regression and learned that you don’t need a white coat to operate a computer.
Dr. John Rutledge
Chief Investment Strategist
Summary: My first device was not a Smartphone, a PC, or a Mac. It was an IBM 1401 about the size of Volkswagen. Here’s how I learned to use it to write my undergraduate Senior Thesis.
As I have written before, I spent the first half of 1968 in West Berlin pretending to go to classes at the Freie Universität von Berlin, sleeping all day, researching the clubs on the Kurfürstendamm all night, and trying, in vain, to get the East German guards on the machine gun towers along the wall to engage in conversation. But when I came back home that summer it was time to get back to work. I had a senior thesis to write.
I had just read Milton Friedman’s “The Role of Monetary Policy”, where he argued there are two things the Fed cannot do: “(1) It cannot peg interest rates for more than very limited periods; (2) It cannot peg the rate of unemployment for more than very limited periods.”
The second claim, of course, unleashed a tsunami of research papers by Phelps, Gordon, Barro, Sargent, Lucas, and others specifying and estimating expectations-augmented Phillips Curves. That led to Rational Expectations, Real Business Cycles, and the Dynamic Stochastic General Equilibrium (DSGE) models that are used today by all the major central banks. (I am not a fan.)
I was more interested in Friedman’s first claim, that the Fed does not have the power to set interest rates. I had learned in my economics courses that increasing the money supply pushes interest rates down, known as the Liquidity Effect. But I had also learned about the Fisher Effect, that increasing the money supply increases inflation and pushes interest rates up. I thought it might be interesting to find out which of these two seemingly conflicting arguments showed up in real world data. (Some years later, I found myself in Milton Friedman’s Money Workshop presenting a paper on the topic to the assembled illuminati. I wrote about it here.)
My First Computer
When I explained to my Senior Thesis advisor what I wanted to write about, he said “You’re going to need a computer to do that. I think there’s one over in the basement of the administration building.”
I knew what a computer was, but I had never seen one. I expected flashing lights, whirring tape drives, and operators in long white coats. What I found was a basement filled with dusty equipment and no one there to run it. The college had installed the computer to run payroll, but abandoned the idea when they learned it was cheaper and easier to do the work by hand. Lucky me!
Figure 2: IBM 1401 components. (Ours didn’t have tape drives or people.)
I turned on the light, found a 1401 Reference Manual in one of the console drawers, and started on page 1. By the time I finished the manual, I knew how to turn on the Card Read Punch, the Console, the (8K) CPU, and the Line Printer, shown above, and how to punch a lot of buttons.
Unfortunately, there was still a long list of things I didn’t know how to do. I knew how to describe my topic in words but didn’t know enough monetary theory to present the analysis or specify the equations. I had read enough macroeconomics articles to know what a regression equation was but hadn’t had a course in statistics or econometrics yet so I didn’t know how to estimate one or how to interpret the results. And I knew I could find data on the money supply and interest rates in the library but had no idea how to stuff them into one end of a machine that could only read 0’s and 1’s and get regression estimates out the other end. But I had the whole year to do it and there was a library on campus with a lot of books. How hard could it be?
Operating the computer was very complicated. Basically, it was just a device that could keep track of a series of simple arithmetic calculations (plus, minus, multiply, divide) using instructions (code) that I had to give it ahead of time to tell it what to do. But the computer could only communicate in machine language (0’s and 1’s) and I, alas, could only speak words. Here are some of the things I had to learn in order to do the work.
First, Monetary Theory
Since my purpose was to find out whether an increase in the money supply made interest rates go up or down, I first needed to know what economists had written about it. So, I went to the library, sat down on the floor in front of Dewey Decimal System section 339 (macroeconomics and related topics), and read my way from the left-hand side of the top shelf to the right-hand side of the bottom shelf. Fortunately—another story for another time—I was a fast reader. When I finished everything in section 339—Samuelson, Solow, Friedman, Keynes, Hayek, Hicks, Fisher, Wicksell, Böhm-Bawerk, Allais, etc.—I stopped.
Most of the things I section 339 had been written since the Great Depression of the 1930’s and focused on whether increasing the money supply would lower interest rates and stimulate sufficient spending to raise output and employment. According to that literature money supply and interest rates should be negatively correlated.
In contrast, Irving Fisher’s work written before the Great Depression (Appreciation and Interest (1896), The Rate of Interest (1907), The Purchasing Power of Money (1911), and The Theory of Interest (1930)) developed the idea that sustained increases in the money supply would produce inflation, erode the purchasing power of money, and drive interest rates higher. As a result, over sufficiently long periods, the money supply and interest rates should be positively correlated.
To test the idea, Fisher used computers—undergraduate students armed with pencils and ledger pads—to estimate a regression equation.
Now I had a theory to test; I just didn’t know how to test it.
Next, Regression Analysis
At that point, it was clear that I needed to estimate the regression equation:
r = a + bM
where r is the interest rate and M is the money supply so I could find out whether the estimate of “b” was positive (Fisher Effect) or negative (Liquidity Effect). So, I found a statistics book in section 519 (probabilities and applied mathematics), started on page 1, and learned how to do the calculations to estimate a regression equation.
Next, Find the Data
To estimate the regression, I needed data on interest rates and money supply. There were no online data bases at that time—in fact, there was no online at all—so I got out my pencil and copied the monthly figures from successive issues of the Federal Reserve Bulletin for a number of previous years (can’t remember how many) onto a ledger pad.
Next, Learn FORTRAN and Write a Regression Program
There was no regression program available at my college so I had to write one. I found another book to learn how to write programs in FORTRAN II. I wrote a program to make all of the calculations I needed to make to estimate the regression equation and to instruct the machine how to send the results to the Line Printer.
Next, Build the Card Deck
Back in the basement, it was time to put the pieces together. I learned how to use the Card Read Punch to punch each line of the FORTRAN program I had written and each observation of the data I had collected onto separate punch cards. Now I had a stack of hundreds of cards six inches high.
Next, Load the FORTRAN II Compiler
Now, it was time to run the programs. To do that, I had to assemble a huge stack of punch cards. First, I needed a couple of special header cards to boot the system and tell the machine to wake up and pay attention. Then came a stack of cards called a FORTRAN II Compiler that I found in a cupboard. The compiler’s job was to instruct the CPU how to translate the FORTRAN II program I had written in letters, spaces, and numbers on cards into machine language (0’s and 1’s), so the computer could read it. Then I needed a couple more header cards followed by the punch cards for the FORTRAN II program I had written to make the regression calculations, then a stack of cards containing the data, then more cards instructing the machine to send the results of the calculations to the printer and, finally, two more header cards to sell the computer it could shot down and take the rest of the day off.
Next, Run the Program
Finally, I loaded the big stack of cards I had prepared into the card reader and pushed the GO button. I remember the excitement of hearing the chunk, chunk, chunk as the cards went through the reader. Then, … nothing. The computer had choked on the first of many SYNTAX ERRORS I had made by misspelling a command, by forgetting a space, or by punching something in the wrong column on a single card. Each time it found a syntax error the whole thing stopped, I had to punch a new card without the syntax error and start the whole GO button thing over again.
Finally, Read and Interpret the output
I remember being shocked when, the printer began to type out lines of text showing the program code, the data, and the regression results, including the estimate of b that was negative and about three times the size of its standard error. The Liquidity Effect had shown up in the data!
Then it dawned on me that, after months of work, I still needed to write the doggone thesis. Bummer.
Lessons for Later
Later, of course, I learned enough to know that the simple procedures I used were not capable of properly evaluating the impact of changes in the money supply on interest rates. For one thing, I had not provided a way to measure the impacts of changes in money supply on interest rates over time, which was important because the Liquidity Effect should take place quite quickly while the Fisher Effect would be likely to show up gradually over a long period of time. But I felt good that I had fought my way through so many obstacles to achieve a result at all.
Then I read that Irving Fisher had obtained his regression results with no machine at all by setting up a series of tables in the Yale gymnasium. The tables were connected by masking tape on the gymnasium floor to tell the students who to hand their results to next in the sequence of calculations. Each table was occupied by two students (he called them “computers”) armed with pencils and ledger pads, both making the same calculation (as a check on accuracy) and passing on their result to the pair of students at the next table until reaching a final result at the last table.
Today, of course, everything is different. References and databases are available online and laptop computers can estimate hundreds of regressions and draw charts showing their results using software like Microsoft Excel, Stata, or MatLab in a heartbeat. But nothing on the screen of a laptop will ever be quite as exciting as the sound of that first “chunk, chunk, chunk,…print” that I heard in 1968.
And I learned an important lesson that paid off many times later in life. If you have enough time, you can learn anything. And you can fight your way through any problem by breaking it down into pieces and attacking each piece in succession until you reach the finish line. You just have to be keep plowing ahead and do the work along the way.
Dr. John
The views and opinions expressed in this article are those of Dr. John Rutledge. Assumptions made in the analysis are not reflective of the position of any entity other than Dr. Rutledge’s. The information contained in this document does not constitute a solicitation, offer or recommendation to purchase or sell any particular security or investment product, or to engage in any particular strategy or in any transaction. You should not rely on any information contained herein in making a decision with respect to an investment. You should not construe the contents of this document as legal, business or tax advice and should consult with your own attorney, business advisor and tax advisor as to the legal, business, tax and related matters related hereto.