ECONOMIC METHODOLOGY
WEEK 1
CHAPTER 1
- Ontology → how reality is formed
- Epistemology → theory of knowledge
- Method → techniques how to get knowledge
- Sociology of science → what scientists do
- Positive science → facts (describing, predicting, explaining)
- Normative science → norms (prescriptions, recommendations)
History – the Wiener Kreis
- Sticking to the facts
- Science = meaningful
o Economically verifiable – synthetic a posteriori (by observation)
o Logically verifiable – analytic (by logical analysis)
o Not meaningful – synthetic a priori (cannot be verified)
Two foundations of logical empiricism
- Logical analysis → formal mathematic term
- Empiricism → observation and measurement (observable in practice)
- 2 steps
o 1. Statement has to be empirically verifiable
o 2. Statement has been verified
- Operationalisation → defining theory in observable terms, theoretical concept =
definitions of observable theories
o Different ways of measuring unemployment
- Demarcation → dividing between scientific and non-scientific
o Scientific if – logically consistent, operationalised
- Context of discovery (irrelevant) → how a law was discovered
- Context of justification → the way it is verified
- Logical positivists can only prove things but never disprove
Symmetry thesis → scientific predictions and scientific explanations meet the same
requirements of the deductive nomological model (DN) → explanation and prediction have
the same logical structure
Deductive nomological model – Hampel
- Explanans
o 1. Lawlike proposition
o 2. Conditions
- Explanandum
, o 3. Phenomenon to explain
- Symmetry thesis
o Explaining (3) requires finding (1) and (2)
o If (1) and (2) → predicting (3)
▪ Domain is not limited by finite space, time, group
- Example:
o Law → quantity theory: in the long run the growth rate of the nominal money
supply is equal to the growth rate of nominal GDP
o Conditions → In the US the annualized growth rate of M1 between 1971 and
2017 was 6%
o Prediction → in the US the annualized growth rate of nominal GDP between
1971 and 2017 was 6%
Humean problem of induction
- Laws are not scientific (black swans)
o Scientific – can be observed
o Non-scientific – not checkable
- Instrumentalism (Schlick)
o Universal statement ≠ meaningful
o Meaningful = certain
o Laws are not scientific
o Laws judged by usefulness to explain
o Universal statements are not scientific
- Confirmationism (Carnap)
o Probability instead of certainty
o Degree of confirmation (x%)
o Laws are scientific but never certain
o Universal statements are scientific but with a certain probability
Why is a law subjective?
- 1. Not clear how to define what is a scientific law → semantically verified
unrestricted generalisation (not accidental)
- 2. Operationalism → how theoretical concepts can be measured (but it can be done
in different ways)
- 3. Hume’s problem of induction → induction derives conclusions from experiences
and observations (somewhat subjective)
- Problem
o Scientific laws require unverified background theory (theoretical terms cannot
be operationalised)
o Scientific theories are never completely neutral
- Logical positivism has failed
- Scientific requirements (meaning) → gain meaning through operationalization
- Syntactic requirements (form) → for generalisation to be law it needs to be
unrestricted generalisation (not restricted to space/time)
,CHAPTER 2
Methodologies of positive economics
- Econometrics – Tinbergen’s approach
o 1. Model must be specified in explicit and functional form (linear)
o 2. Decide appropriate data definitions (operationalization)
o 3. Bridge between theory and data (verify key stats)
Keynes-Tinbergen debate (Keynes’ criticism of Tinbergen) → is it going to hold for the
future?
- Tinbergen’s method → a lot of complicated equations (mathematical model)
- Criticism by Keynes
o Not testing just measuring
o Measurements are only correct with complete model (all causalities are
known)
- 3 points of scepticism
o 1. Omitted variable bias → estimates of the causal factors are biased unless
all causal factors are taken into account – impossible
o 2. Data mining → too much trial and error to make sure the data fits the
equations
o 3. Invariance problems → is it going to hold in the future
Haavelmo – Probability approach
- We can continue measuring → if model is complete enough → law
- Problem of passive observations → we do not know whether the variable has
potential influences in the future or not
- The problem is that in economic variables are strongly interdependent with many
variables involved and we cannot do controlled experiments
o Variable may not be significant now → but might be after certain threshold
o If we wait → model is incomplete
o If we include everything → model is too complex
o Keynes is right → we can only measure
Measurement-without-theory debate
- Cowles Commission (Koopmans) vs NBER (Vinning)
- Koopmans criticised NBER → theory is based on observations (tells us what is
important)
- NBER → descriptive statistics, not model based, inductive → no observations
- Cowles Commission → empirical research based on theory
o 1. A priori math model
o 2. Estimated with simultaneous equations econometrics
o 3. Hypothesis testing
- Koopmans – arguments against measurement without theory
, o There are many variables and relations that may play a role that we need
theory to know which relations have to be tested empirically in a meaningful
way
o Theory is needed to tell us which relations are time invariant
o Statistical analysis presupposes assumptions which cannot be tested with the
same data
o Theory is needed to clarify how the empirical research have to be interpreted
o Without theory it is not clear what a researcher means by the applied
concepts and terms
- Response of NBER – Vinning
o Cowles Commission’s model is ‘pretty skinny felow’ → too simple to capture
the diversity of data
o Inductive approach – descriptive statistics is useful to discover hypothesis
Problems of econometrics
- No lab
- More variables → more certainty but too complex
- Starting big (NBER x Tinbergen) x start small theoretical basis and expand (Koopmans
x Cowles C, Haavelmo)
- Complexity unavoidable
Methodology of positive economics – Friedman
- Dealing with complexity – critical of large mathematical models
- Friedman’s arguments against macroeconomic models
o Too much trial and error → invariance problem
- Naïve models work better
o 𝑦𝑡+1 = 𝑦𝑡 + 𝜀𝑡
o 𝑦𝑡+1 = 𝑦𝑡 + (𝑦𝑡 − 𝑦𝑡−1 ) + 𝜀𝑡
o Marshal → simple model first
o Friedman → model should predict well
- A model is scientific if it predicts well and in well-specified domains
- We should start with small parts (models) and later combine them into one large
macroeconomic model
Interpretation of Friedman:
- 1. In favour of economic instrumentalism
o Whether assumptions are realistic or not does not matter as long as they
predict well
o More radical than Schlick
- 2. Anti-realisticness – Friedman opposing realism
o Realism → all lawlike statements are considered scientific (no matter how
bad the correlation is)
o Realism of assumptions does not matter
o Problem: anything goes
WEEK 1
CHAPTER 1
- Ontology → how reality is formed
- Epistemology → theory of knowledge
- Method → techniques how to get knowledge
- Sociology of science → what scientists do
- Positive science → facts (describing, predicting, explaining)
- Normative science → norms (prescriptions, recommendations)
History – the Wiener Kreis
- Sticking to the facts
- Science = meaningful
o Economically verifiable – synthetic a posteriori (by observation)
o Logically verifiable – analytic (by logical analysis)
o Not meaningful – synthetic a priori (cannot be verified)
Two foundations of logical empiricism
- Logical analysis → formal mathematic term
- Empiricism → observation and measurement (observable in practice)
- 2 steps
o 1. Statement has to be empirically verifiable
o 2. Statement has been verified
- Operationalisation → defining theory in observable terms, theoretical concept =
definitions of observable theories
o Different ways of measuring unemployment
- Demarcation → dividing between scientific and non-scientific
o Scientific if – logically consistent, operationalised
- Context of discovery (irrelevant) → how a law was discovered
- Context of justification → the way it is verified
- Logical positivists can only prove things but never disprove
Symmetry thesis → scientific predictions and scientific explanations meet the same
requirements of the deductive nomological model (DN) → explanation and prediction have
the same logical structure
Deductive nomological model – Hampel
- Explanans
o 1. Lawlike proposition
o 2. Conditions
- Explanandum
, o 3. Phenomenon to explain
- Symmetry thesis
o Explaining (3) requires finding (1) and (2)
o If (1) and (2) → predicting (3)
▪ Domain is not limited by finite space, time, group
- Example:
o Law → quantity theory: in the long run the growth rate of the nominal money
supply is equal to the growth rate of nominal GDP
o Conditions → In the US the annualized growth rate of M1 between 1971 and
2017 was 6%
o Prediction → in the US the annualized growth rate of nominal GDP between
1971 and 2017 was 6%
Humean problem of induction
- Laws are not scientific (black swans)
o Scientific – can be observed
o Non-scientific – not checkable
- Instrumentalism (Schlick)
o Universal statement ≠ meaningful
o Meaningful = certain
o Laws are not scientific
o Laws judged by usefulness to explain
o Universal statements are not scientific
- Confirmationism (Carnap)
o Probability instead of certainty
o Degree of confirmation (x%)
o Laws are scientific but never certain
o Universal statements are scientific but with a certain probability
Why is a law subjective?
- 1. Not clear how to define what is a scientific law → semantically verified
unrestricted generalisation (not accidental)
- 2. Operationalism → how theoretical concepts can be measured (but it can be done
in different ways)
- 3. Hume’s problem of induction → induction derives conclusions from experiences
and observations (somewhat subjective)
- Problem
o Scientific laws require unverified background theory (theoretical terms cannot
be operationalised)
o Scientific theories are never completely neutral
- Logical positivism has failed
- Scientific requirements (meaning) → gain meaning through operationalization
- Syntactic requirements (form) → for generalisation to be law it needs to be
unrestricted generalisation (not restricted to space/time)
,CHAPTER 2
Methodologies of positive economics
- Econometrics – Tinbergen’s approach
o 1. Model must be specified in explicit and functional form (linear)
o 2. Decide appropriate data definitions (operationalization)
o 3. Bridge between theory and data (verify key stats)
Keynes-Tinbergen debate (Keynes’ criticism of Tinbergen) → is it going to hold for the
future?
- Tinbergen’s method → a lot of complicated equations (mathematical model)
- Criticism by Keynes
o Not testing just measuring
o Measurements are only correct with complete model (all causalities are
known)
- 3 points of scepticism
o 1. Omitted variable bias → estimates of the causal factors are biased unless
all causal factors are taken into account – impossible
o 2. Data mining → too much trial and error to make sure the data fits the
equations
o 3. Invariance problems → is it going to hold in the future
Haavelmo – Probability approach
- We can continue measuring → if model is complete enough → law
- Problem of passive observations → we do not know whether the variable has
potential influences in the future or not
- The problem is that in economic variables are strongly interdependent with many
variables involved and we cannot do controlled experiments
o Variable may not be significant now → but might be after certain threshold
o If we wait → model is incomplete
o If we include everything → model is too complex
o Keynes is right → we can only measure
Measurement-without-theory debate
- Cowles Commission (Koopmans) vs NBER (Vinning)
- Koopmans criticised NBER → theory is based on observations (tells us what is
important)
- NBER → descriptive statistics, not model based, inductive → no observations
- Cowles Commission → empirical research based on theory
o 1. A priori math model
o 2. Estimated with simultaneous equations econometrics
o 3. Hypothesis testing
- Koopmans – arguments against measurement without theory
, o There are many variables and relations that may play a role that we need
theory to know which relations have to be tested empirically in a meaningful
way
o Theory is needed to tell us which relations are time invariant
o Statistical analysis presupposes assumptions which cannot be tested with the
same data
o Theory is needed to clarify how the empirical research have to be interpreted
o Without theory it is not clear what a researcher means by the applied
concepts and terms
- Response of NBER – Vinning
o Cowles Commission’s model is ‘pretty skinny felow’ → too simple to capture
the diversity of data
o Inductive approach – descriptive statistics is useful to discover hypothesis
Problems of econometrics
- No lab
- More variables → more certainty but too complex
- Starting big (NBER x Tinbergen) x start small theoretical basis and expand (Koopmans
x Cowles C, Haavelmo)
- Complexity unavoidable
Methodology of positive economics – Friedman
- Dealing with complexity – critical of large mathematical models
- Friedman’s arguments against macroeconomic models
o Too much trial and error → invariance problem
- Naïve models work better
o 𝑦𝑡+1 = 𝑦𝑡 + 𝜀𝑡
o 𝑦𝑡+1 = 𝑦𝑡 + (𝑦𝑡 − 𝑦𝑡−1 ) + 𝜀𝑡
o Marshal → simple model first
o Friedman → model should predict well
- A model is scientific if it predicts well and in well-specified domains
- We should start with small parts (models) and later combine them into one large
macroeconomic model
Interpretation of Friedman:
- 1. In favour of economic instrumentalism
o Whether assumptions are realistic or not does not matter as long as they
predict well
o More radical than Schlick
- 2. Anti-realisticness – Friedman opposing realism
o Realism → all lawlike statements are considered scientific (no matter how
bad the correlation is)
o Realism of assumptions does not matter
o Problem: anything goes
