The Militarized Zone between Theory and Practice in Economics
Judy L Klein False Dichotomies: CIGI INET Conference, Waterloo, ON, November 17, 2012
Dichotomies in Economics
Theory Practice
Pure Applied
Positive Normative
Descriptive Prescriptive
Military Planning Contracts 1940-1960 Valorized
↓
↓ ← Practice for military client fed back into theory, new models, modeling strategies
↑ Technical revolution of
new modeling strategies • Erosion of micro / macro boundary • Erection of new boundaries:
theory/practice, economics/economy
Effect on macroeconomic theory of modeling for US military client
agent modeled as collection of decision rules
“benevolent social planner” is invisible hand
communism of models
recursive dynamics w/uncertainty
macro monoculture of modeling strategizing
Need to reduce flash-to-bang time
Capacity for mass production (effective computation) by machines/unskilled labor
Capacity for mass operation by unskilled labor
Military Patrons insisted model design process
had to take into account:
Common Features of Mathematical Protocols for Military
• Solution is a quantified policy , a decision rule of action
• Computability - effective numerical solutions:
o approximate rather than exact solutions
o use constraints of situation to ensure convergence
o solve problems in policy space
o coding protocol as a recursive algorithm
• Allocation criterion minimizing losses or maximizing gain
• Incorporation of uncertainty
• Conditional, sequential, decision process
Sequential Analysis: January 12, 1943 Capt. Garrett Lansing Schuyler, research head of USN Bureau of Ordnance asks Allen Wallis, Director of Research of SRG-C?
how to
• determine whether new experimental gun is superior to an old gun
• determine number of trials necessary for adequate statistical comparison of the percentage of hits with two different guns
Wallis memo to Schuyler “Statistical Analysis of Double Dichotomies” 01/18/43
Early March 1943, Schuyler suggests control chart
Inspector continuously plots results
• If curve of results crosses one line:
o halt experiment
o new gun is superior
• If results crosses other line:
o halt experiment
o old gun is superior
Chart from Lt. Col. Leslie Simon, Army Aberdeen Proving Ground, 1942
The job of inspection is twofold: the inspector must be technically competent to judge the job, and he must also be able to predict lot quality from samples. He cannot be expected to know probability theory, so he cannot make predictions unless procedure is boiled down to rules. Conrad, US Navy 1944
Sequential Analysis: Abraham Wald, Theorem Chief AMP 30.1 September 1943
SRG Team Application Mathematicians AMP30.2R, Sept. 1945
Division of Labor in Modeling Production Process
Sequential Analysis
1943-1944
Theorem managers Wald (1943) devises
log probability ratio test
Application mathematicians
SRG team constructs
equations for graphical templates
nomograms
narrates examples
Mid-level production managers Use nomograms, slope & intercept equations to
draw chart decision boundary lines
Unskilled labor Enters data on chart to
update decision rule
Division of Labor in Modeling Production Process Dynamic Programming w/ Quadratic Costs
1952-1954
Theorem managers Bellman (1949) principle of optimality
Simon (1954) certainty Equivalent Theorem
Application mathematicians
“Paint Factory” team at Carnegie Institute
constructs linear decision rule templates,
narrates examples
Mid-level production
managers
Use desk calculators to
specify parameters of linear decision rule
Unskilled labor Enters data on desk calculator to
update decision rule
Dynamic Programming- RAND 1950’s optimal allocation of resources among time periods
w/ approximations in policy space
By a solution we mean a set of rules which tell us which operation to perform at every stage in every situation.
Dynamic Programming Richard Bellman’s Gold Mining with Atomic Weapons (RAND 1951)
employed ispolicy optimalan and , has , has when
damaged is machine thebefore mined gold ofamount expected),(
yBxAyxf (1)
),(),,(max),( yxfyxfyxf ba (2)
….
yrxfyrpB
yxrfxrpAyxf
222
111
1,:
,1:max),( (5)
Bellman’s solution was at each stage:
optimal is choice either ,)1()1( if
choice the take,)1()1( if
choice, the take,)1()1( if
222111
222111
222111
pyrppxrp
Bpyrppxrp
Apyrppxrp
(6)
At each stage choose the mine (the target) with highest ratio of immediate expected gain to expected loss.
Glory to the conquerors of the universe!
October opened the road to space! http://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpg
Soviet Report- Oct. 4, 1957
http://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpghttp://cdn2.retronaut.co/wp-content/uploads/2012/02/October-opened-the-road-to-space.jpg
calculus of variations: curve = locus of points;
dynamic programming: curve = envelope of tangents
approximations in policy space
policy space “domain for activists”
Time,
hr.
Observation Velocity
correction
mph
Miles to
target
Indicated
velocity
correction
Uncertainty in
velocity correction
mph
0.6 Moon, Antares 2528 0 11.9
0.9 Earth Fomalhaut 1504 1.3 12.9
1.2 Earth, Deneb 1404 11.0 9.2
1.5 Earth, Aldebaran 1342 12.6 9.2
1.8 Earth, Aldebaran 1290 14.0 9.2
2.2 Earth, Aldebaran 1224 15.5 9.4
2.6 Earth, Pollux 1136 16.9 9.5
3.0 Earth, Procyon 1013 18.3 9.2
3.4 Earth, Procyon 928 19.7 8.7
3.8 Earth, Pollux 825 20.9 8.3
4.5 Earth, Procyon 719 22.8 8.0
5.0 24.1
5.5 Earth, Pollux 573 0 7.6
6.0 Earth, Procyon 504 4.6 6.4
6.5 Earth, Pollux 441 5.7 5.8
7.0 Moon, Antares 395 6.5 5.3
IBM 305 RAMAC the first ‘SUPER’ computer with a hard disk drive that weighed over a ton and stored 5 MB of data. 1957 - Texomatube
Libratrac-2000 airborne computer stored 2048 words, weighted 32 pounds 1959
http://www.texomatube.net/video/play/hard__disk_drive_back_in__1956_With_5__MB_of__storage
Kalman Filter protocol for correcting trajectories
Developing framework for broad optimal allocation problems for the USAF & Navy:
• Generalizable mathematical model
• Algorithm as solution to mathematical problem Kenneth Arrow- “algorithm as capital”
• Rational procedure for designing model & algorithm for scarce computational resources Herbert Simon- “procedural rationality” Thomas Sargent- “modeling strategy”
Herbert Simon: “This particular blend of fundamental social science and mathematics with business applications; of economics with behavioral science and organization theory is more and more being identified by colleagues at other institutions with the ‘Carnegie group’.”
Graduate School of Industrial Administration Carnegie Institute of Technology
William Cooper:
“Applications-driven theory”
Key Dates Carnegie Institute of Technoloby – GSIA
1949 Project SCOOP 3-yr research grant – Intra-Firm Planning and Control
1952 Office of Naval Research- grant Planning and Control of Industrial Operations
1952-54 Simon, Holt , Modigliani formulate dynamic programming problem as Quadratic cost minimizing objective function & Linear decision rules
1953 In RAND memo- Simon proposes model describing “rational choice by organisms of limited computational capacity.”
1957 Simon’s articulation of Bounded Rationality
1960 Muth’s Rational Expectations = equilibrium modeling strategy
1966 Lucas’s CIT WP on “Optimal Investment with Rational Expectations”
1973 Simon: Substantive Rationality (Economist’s focus on outcome) Procedural Rationality (Psychologists' focus on process)
1976
Lucas, "Econometric Policy Evaluation: A Critique," Carnegie-Rochester Conference Series on Public Policy
For Holt, Modigliani, Muth, and Simon working on the ONR Paint Factory Project “uncertainty became our common bond.”
With quadratic cost function • Parameters of rules
computed w/ desk calculator in 3 man-hours
• Computation of monthly decisions w/ 5-minute calculation
(Holt, Modigliani, and Simon 1955, 2)
With Quadratic Cost Function for Dynamic Programming “planning problem for the case of uncertainty can be
reduced to the problem for the case of certainty” Expected Value of price = Certain price
Simon 1956
Holt, Modigliani, Muth, and Simon 1955
There should be no impossible gulf between “theory” and “practice”: good theory is theory that works. A theory that tells you that if you knew a man’s utility function, or if you could forecast sales perfectly, you could make an optimal or rational decision gives little comfort—unless the numbers that have to be substituted in the equations are available or can be supplied. A theory of rational decisions is a theory of how to decide, given certain kinds of information and certain computing capacities. (Simon and Holt, ONR Memo, 1954, 16-17)
Quadratic Cost Function for Dynamic Programming
Bounded Rationality Replacement of goal of maximizing with goal of satisficing
Muth’s Rational Expectations: Certain price = expected price = equilibrium price Econometric Society Washington DC Meeting, December 1959
Simon’s Notes for talk in Sweden 1975
Equilibrium Modeling Strategy is Procedural Rationality
Muth 1961
1966
Key Dates Graduate School for Industrial Administration (GSIA)-CIT
1949 Project SCOOP 3-yr research grant for Intra-Firm Planning and Control
Fall 1952 Office of Naval Research-Planning and Control of Industrial Operations
1953-54 Simon, Holt , Modigliani formulate dynamic programming problem as Quadratic cost minimizing objective function & Linear decision rules
1953 In RAND memo- Simon proposes model describing “rational choice by organisms of limited computational capacity.”
1957 Simon’s articulation of Bounded Rationality
1960 Muth’s Rational Expectations = equilibrium modeling strategy
1966 Lucas’s CIT WP on “Optimal Investment with Rational Expectations”
1973 Simon: Substantive Rationality (Economist’s focus on outcome) Procedural Rationality (Psychologists' focus on process)
1976 Lucas, "Econometric Policy Evaluation: A Critique," Carnegie-Rochester Conference Series on Public Policy
They are at the same time the most refractory to the accepted techniques of mass attack and the division of labor.
http://www.cigionline.org/sites/default/files/shared/banner.jpg
Norbert Wiener 1946
It is these boundary regions of science, which offer the richest opportunities to the qualified investigator.
Effect of wartime modeling on US macroeconomic theory?
agent modeled as collection of decision rules
“benevolent social planner” metaphor replaces invisible hand
recursive dynamics w/uncertainty
abiding modeling strategies
• Quadratic loss function- certainty equivalence: expected price = certain price
• Rational expectations = communism of models expected price = equilibrium price
Macro Monoculture of Modeling Strategizing
Technical revolution in Economics Monoculture of Modeling Strategy
Research in rational expectations and dynamic macroeconomics has a momentum of its own. That momentum stems from the logical structure of rational expectations as a modeling strategy….. It’s more a technical revolution. …These are technical issues about staring at models. Sargent 1982
New classical economics means a modeling strategy. Townsend, 1983
While rational expectations is often thought of as a school of economic thought, it is better regarded as a ubiquitous modeling technique used widely throughout economics. Sargent, 2008
Modeling turn- Linear Programming Alex Orden 1993,
Robert E. Lucas 1979” “Rational expectations” is a term John Muth coined to refer to a model-building principle …. It is a property which a mathematically explicit economic model either does or does not possess. One can ask, for example, whether expectations are rational in the Klein-Goldberger model of the United States economy; one cannot ask whether people in the United States have rational expectations….
The limited intelligibility of popular discussions of rational expectations is, I think, entirely due to the understandable desire to evade this central fact.
“To put it more bluntly, control theory does not deal with the real world, but only with mathematical models of certain aspects of the real world; therefore the tools as well as results of control theory are mathematical. There is a close analogy between this situation and the evolution of probability theory into a strictly mathematical discipline. Kalman, R.E.; Falb, P.L.; Arbib, M.A., 1969
Modeling Turn- Macroeconomics & Control Engineering
Notes from NBER Rational Expectations Conference March 21-22 1974, Ben Friedman: there is a straight-forward explanation for the ability of models incorporating rational expectations to yield classical results. Specifically, the statistical assumptions about information structure implied by the rational expectations hypothesis impart to the analysis a long-run equilibrium nature … Franco Modigliani :The technical results appear valid, given the assumptions. However, he questioned a couple of its assumptions: a vertical Phillips curve and the extreme form of rational expectations. He particularly doubted that individuals could solve this “phenomenal dynamic problem.” Muth objected that what Modigliani refers to as a disadvantage is actually an advantage: the extreme assumptions provide a structure for our model which gets results.
Military needs induced a mathematical construction of rationality while simultaneously binding that rationality with computational reality. Modeling strategy ensued and eventually became a key driving force to the development of economic theory
• Applied mathematics becomes a science of economizing
• Economics acquires mathematical capacity for solving dynamic allocation problems w/uncertainty
• New “explicit methodology” for applied mathematicians
o “building of an adequate mathematical model as the fundamental step…and, indeed the only inductive or creative step.”
o “solution may simply be an algorithm”
o model “design itself is a statistical decision. Rationally it demands measurements “
Assistant Secretary of the Air Force for Research and Development, Brockway McMillan 1957, 1961
• Rationalization of mathematical protocol production process
Key Transformations- 1943-1976
Unmet Needs w/ high significance
• How to model uncertainty, including expectations
• How to model cycles & reconcile w/ concept of eqiulibrium
• New fodder for debate over free markets vs government intervention
• Way of unifying micro foundations with macroeconomic theory
Division of Labor in Modeling Production Process Sequential Analysis
1943-1944
Dynamic Programming w/
Quadratic Costs 1952-1954
Theorem managers Wald (1943)
log probability ratio
Bellman (1949)
principle of optimality
Simon (1954) certainty
Equivalent Theorem
Application
mathematicians
SRG team constructs
nomograms
equations for graphical templates
narrates examples
GSIA team constructs
linear decision rule templates
narrates examples
Mid-level production
managers
Use nomograms, slope & intercept
equations to draw chart decision
boundary lines
Use desk calculator to specify
parameters of linear decision
rule
Unskilled labor Enters data on chart to
update decision rule
Enters data on desk calculator to
update decision rule
Effect of Cold War modeling on US macroeconomic theory?
Benevolent planner- Stokey, Lucas, and Prescott (1989)
“there is a wide class of situations in which the “invisible hand” ensures that the sets of Pareto-optimal allocations and competitive equilibrium allocations coincide exactly. In these situations we can interpret certain normative models of optimal decision-making (from the point of view of a hypothetical “benevolent social planner”) as positive models of equilibrium outcomes.”
Individual as collection of decision rules- Lucas 1986
we view or model an individual as a collection of decision rules (rules that dictate the action to be taken in given situations)
Communism of models as a modeling strategy: Sargent 2007
“All Agents inside the model, the econometrician, and God share the same model. The powerful and useful empirical implications of rational expectations—the cross-equation restrictions and the legitimacy of the appeal to a law of large numbers in GMM [Generalized Method of Moments] estimation—derive from that communism of models.”
Fundamental Study of Adaptive Control Systems R. E. Kalman, T. S. Engiar, R. S. Bucy, RIAS The Martin Company, April 1962
Simon’s Reverse Engineering to derive Quadratic Cost Function
The enemy is constantly improving, and we must therefore continuously develop to maintain superiority. Superiority in time is the important objective. …[Sperry’s] responsibility is to find the problem, develop the solution, prove it, design it for production and accomplish its initial production, all in a minimum of time… Time is the essence of our problem; me must maintain superiority; CEO Gillmor, Sperry Gryoscope CEO 1943
General Hap Arnold, Aircraft Procurement Conference 1939
It is going to cost us money, but we want to eliminate time element.
Shorter Flash-to-Bang
Mathematical model for ball-cage integrator = adaptive expectations
Sperry Gyroscope Company K-series Lead Computing Sight
April 1940- US AAF contract for 1 pilot model
Designed for mass production mass operation
Lead Computing Problem
Mass Production of Sperry Computing Gun sights by unskilled labor
Mass Operation: Rules of action were the solution to lead computing problem
AAF would only devote 3 days training for gunners
New avg. angular velocity estimate = β (new velocity input) + (1- β) (old avg. velocity estimate)
with β symbolizing a fractional value between zero and one
The expected rate of change in prices is
revised per period of time in proportion to the
difference between the actual rate of change in
prices and the rate of change that was
expected.
Cagan 1956
10,ˆ)1(ˆ 1 ttt yyy