News, Stock Prices and Economic Fluctuations Paul Beaudry & Franck Portier University of British Columbia & Universit´ e de Toulouse March 2004 Oxford Anglo-French Meeting 1
Introduction • What drives business cycle fluctuations? • What is the relative importance of demand versus supply shocks? • Can we identify the class of models most capable of explaining the reaction of the economy to such shocks? • We present evidence suggesting that business cycle fluctuations may be primarily (or at least largely) driven by a shock which is neither a traditional demand or technology shock, but is instead a type of hybrid which admits a simple structural interpretation as a news shock. 2
Empirical Strategy • We perform two different orthogonalization schemes as a means of identifying properties of the data, that can then be used to evaluate different theories of business cycles. • We impose sequentially, not simultaneously, either impact or long run restrictions on the orthogonalized moving average representation of the data. • The primary system of variables that interests us is one composed of measured total factor productivity (TFP) and an index of stock market value (SP). • Stock prices are likely to be a good variable for capturing any changes in agents expectations about future economic growth. 3
Main Results • Data on TFP and stock market value have properties that run counter to the demand and supply type dichotomy inherent to most New Keynesian and RBC models. • The innovation in stock prices which is contemporaneously orthog- onal to TFP is actually extremely correlated with the shock that explains long run movements in TFP. • The observed pattern is easily understood as the result of news/diffusion shocks, that is, innovations in agents expectations of future techno- logical opportunities that arise before these opportunities are actually productive in the market. 4
• This particular shock series cause standard business cycle co-movements (i.e., induce positive co-movement between consumption and invest- ment) and explains a large fraction of business cycle fluctuations.
Plan of the talk 1. Using impact and long-run restriction sequentially to learn about macroeconomic fluctuations 2. Data and Specification Issues 3. Results in bi-variate system 4. Higher Dimension Systems 5
1. Using impact and long-run restriction sequentially to learn about macroeconomic fluctuations 1.2. Main Idea • Using short run and long run restrictions in VARs, not simulta- neously, but instead sequentially as a means of evaluating different classes of economic models. 6
• Simple bi-variate system. • Measured total factor productivity TFP t , and a forward looking economic decision variable X t . • The only characteristic of X t that is important for our argument is that it be a jump variable, that is, a variable that can immediately react to changes in information without lag (stock price, interest rate, even consumption). 7
• We consider two alternative representations of the Bivariate VAR with orthogonalized errors: � � � � ∆ TFP t ǫ 1 ,t = Γ( L ) (1) , ∆ X t ǫ 2 ,t � � � � ˜ ∆ TFP t ǫ 1 ,t = ˜ Γ( L ) (2) , ∆ X t ˜ ǫ 2 ,t - short run restriction: Γ 0(1 , 2) = 0 ( ǫ 2 has no contemporaneous im- pact on TFP ) � � = � ∞ -long run restriction: ˜ i =0 ˜ Γ(1) (1 , 2) Γ i = 0 (˜ ǫ 2 has no long run impact on TFP ) 8
• Our idea now is to use these two different ways of organizing the data to help evaluate different classes of economic models. • For example, a particular theory may imply that the correlation between the resulting errors ǫ 2 and ˜ ǫ 1 be close to zero and that their associated impulses responses be different. • Therefore, we can evaluate the relevance of such a theory by ex- amining the validity of its implications along such a dimension. 9
1.2. The predictions of three simple models (a) A New Keynesian model • The model is driven by monetary shocks and surprise changes in technology • It is an economy with no capital, monopolistic competition, mone- tary shocks, pre-set wages and technological disturbances. t − Λ( L j ∞ t ) σ � log C j β t U = E 0 σ t =0 10
�� 1 � 1 ρ 1 , o z ρ 1 y = 0 < ρ 1 < 1 i di �� 1 � 1 ρ 2 , o l ρ 2 z i = θ t 0 < ρ 2 < 1 j dj • θ is a random walk • Firms have a value because there is some monopoly power
• The model solution can be written as � � � � � � ∆ TFP t 1 0 η 1 ,t = (1) ∆ SP t 1 (1 − L ) η 2 ,t • This model implies ǫ 1 = η 1 , ǫ 2 = η 2 , � ǫ 1 = η 1 and � ǫ 2 = η 2 • In particular, this type of model implies that ǫ 2 ⊥ � ǫ 1 . 11
(b) A simple RBC model with technology and preference shocks • TFP is again a random walk � � ( L j t ) σ � ∞ log C j t =0 β t • U = E 0 t − Λ t , Λ t = η 2 iid σ • The model solution can be written as � � � � 1 0 ∆ TFP t η 1 ,t = (2) − (1 − L )(1 − γ ) 2 1 ∆ p b (1 − γ ) 1 − γL − 1 η 2 ,t t σ (1 − γL ) • This model implies ǫ 1 = η 1 , ǫ 2 = η 2 , � ǫ 1 = η 1 and � ǫ 2 = η 2 • In particular, this type of model implies that ǫ 2 ⊥ � ǫ 1 . 12
(c) A model with delayed response of innovation on productivity • Measured TFP, denoted θ , is composed of two components: a non-stationary component D t and a stationary component ν t . = D t + ν t θ t � ∞ = D t i =0 d i η 1 ,t − i 1 − δ i , = 0 ≥ δ < 1 d i = ρν t − 1 + η 2 ,t , 0 ≤ ρ < 1 ν t • We call D t a diffusion process since an innovation η 1 is restricted to have no immediate impact on productive capacity ( d 0 = 0), the effect of the technological innovation on productivity is assumed to grow over time ( d i ≤ d i +1 ) and the long run effect is normalized to 1. 13
• To derive the implied structural moving average for ∆ TFP and ∆ SP , we consider a simple Lucas’ tree type of model • The ownership of the unique tree of the economy is tradable and where it pays dividend θ t . • Households consume and trade firms shares. t =0 β tC 1 − σ � ∞ • U = E 0 1 − σ , σ = 0 t 14
• The model solution can be written as � � (1 − δ ) � ∞ � � (1 − L ) i =1 δ i − 1 L i ∆ TFP t η 1 ,t (1 − ρL ) = � ∞ β (1 − δ ) ∆ SP t βρ 1 − L η 2 ,t i =0 δ i L i 1 − βδ 1 − βρ 1 − ρL 15
• the impact matrix on levels of TFP and SP is of the form: 0 1 (3) β (1 − δ ) βρ 1 − βδ 1 − βρ And the long run matrix for the levels of TFP and SP is of the form � � 1 0 (4) β 0 1 − βδ • This model implies ǫ 1 = η 1 , ǫ 2 = η 2 , � ǫ 1 = η 1 and � ǫ 2 = η 2 • In particular, this type of model implies that ǫ 2 ⊥ � ǫ 1 . • This model implies ǫ 1 = η 2 , ǫ 2 = η 1 , � ǫ 1 = η 1 and � ǫ 2 = η 2 • In particular, we have that ǫ 2 is co-linear to � ǫ 1 . 16
Discussion • The important aspect of the first model is that it implies that business cycle fluctuations that can be decomposed into structurally meaningful supply driven and demand driven components. • The non technological disturbance – which we can refer to as the demand disturbance – should be contemporaneously orthogonal to innovations in TFP and should not cause long run movements in TFP . 17
• The news/diffusion model is different. • Even before technological opportunities have actually expanded an economy’s production possibility set, forward looking variables already incorporate this possibility (implementation cycle models also do the job) • Such a model does not validate a decomposition in terms of demand versus supply effect. • In the short run, an anticipated technological improvement (a news shock) looks like a demand effect, while in the long run it looks like a supply effect. 18
2. Data and Specification Issues 2.1. Data The data we use are U.S. 1948Q1-2000Q4 non farm PBS, per capita • Stock Price: S&P500, deflated by GDP deflator • TFP: Non farm private business sector, computed with GDP, hours, a constant labor share α which is the average over 1948-98, as com- puted by the BLS, capital services (BLS): GDP t log TFP t = log t × Capital services (1 − α ) Hours α t 19
• We check for robustness to various refinements of the TFP measure (variable capital utilization, varying α ) • Consumption is personal consumption of non durable goods and services. • Investment is fixed private domestic investment plus personal con- sumption of durable goods.
TFP Stock Prices % deviation from 1948 % deviation from 1948 250 250 200 200 150 150 100 100 50 50 0 0 1960 1980 2000 1960 1980 2000 Consumption Investment % deviation from 1948 % deviation from 1948 250 250 200 200 150 150 100 100 50 50 20 0 0 1960 1980 2000 1960 1980 2000
2.2. VAR Specification • Main idea: impose as less structure as possible on the data � 6 lags, levels or n − 1 cointegrating relations 3. Bi-variate VAR • We model the joint behavior of TFP and the stock price index. 21
ǫ 2 Against � ǫ 1 in the ( TFP, SP ) VAR, baseline specification 4 3 2 1 ε 1 tilde 0 −1 −2 −3 −4 −4 −3 −2 −1 0 1 2 3 4 ε 2 22
Recommend
More recommend
