András Gregor, Richard Meade, Pavel Svaton with Arnold Salas Toulouse School of Economics, 11 March 2010
Outline y Introduction (Richard) y Context y Motivating examples y The paper’s “Big Idea” and Key Conclusions y Key terms, assumptions and tools y Working through the model – four propositions (Pavel) y Empirical analysis (Andrís) y Discussion (Richard)
Context – Explaining Volatility in GDP Growth Aggregate shocks
• Policy/monetary shocks, inflation, wars, etc
Sectoral shocks
• Shocks affecting specific sectors • Potentially big enough to drive GDP • Represent a weak form of Gabaix’s hypothesis
Firm‐level shocks
• Gabaix is here – arguing that firm‐level shocks can affect GDP fluctuations if the firm‐size distribution has “fat tails”
Motivating Examples y Gabaix mentions companies whose idiosyncratic shocks can
move GDP growth materially:
y Microsoft one‐off dividend in 2004 boosting personal income
growth y Nokia accounting for 1.6% points of Finland’s GDP growth y Strike at GM, innovations by Wal‐Mart, …
y Easy to think of other examples of companies influencing GDP: y Oil companies – Russia (Gazprom 8% of GDP), Hungary, Nigeria, Venezuela, … y Car manufacturers – Italy (Fiat 8.5% of GDP) y New Zealand: y y
Largest company is dairy cooperative, Fonterra (8% of GDP and 25% of exports) Largest listed company is Telecom (≈20% of main stock index)
On the significance of large firms in the US
Æ Sales of top 100 non-oil Compustat firms in US account for 25-30% of GDP
“Big Idea” and Key Conclusions Idiosyncratic shock
Large Firm
Small agg. shocks
Granular hypothesis: Large part of aggregate fluctuations is due to idiosyncratic shocks to individual (large) firms: ÆComplements other explanations ÆLarge firms are “incompressible grains” ÆDemystifies the Solow residual?
GDP
Other Firms (GE)
“Big Idea” (cont’d) y Where does this idea come from (aside from appearing
obvious at first glance, especially to those outside the US)?
y First need to knock down the “straw man”: 1/√N
diversification – i.e. the common argument that aggregate volatility σGDP declines at rate 1/√N as the number N of firms rises, so firm‐level shocks wash out in the aggregate if N is large enough
y Gabaix argues and cites evidence that firm‐size distribution
is fat‐tailed, following a “power law distribution”, and hence σGDP declines slower than 1/√N as N rises Æ firm‐level idiosyncratic shocks don’t just diversify away
Key Findings y Gabaix’s measure of large‐firm idiosyncratic
productivity shocks (“Granular Residual”) explains about ⅓ of GDP growth and the Solow Residual y Idiosyncratic shocks do not die out in the aggregate,
and are large enough to explain GDP fluctuations (more so for non‐US countries which are often less diversified)
Key Terms, Assumptions, Tools y Solow Residual, or Total Factor Productivity (TFP) – the part of
output growth that cannot be explained by increases in inputs (i.e. the empirically important bit) y Granular residual – large‐firm idiosyncratic firm‐level
productivity shock left over after relating labour productivity to total factor productivity and decomposing it into common, industry, and firm‐specific parts Æ see definitions later y Gibrat’s Law of Variances – the standard deviation of a firm’s
percentage growth rate is independent of the firm’s size: y Helps to make his arguments, but turns out to not be essential y Implies firm‐level shocks “scale up” rather than diversify
Key Terms, etc (cont’d) y Power law distribution – e.g. P(S>x) = ax‐ζ y Commonly understood in terms of the “80/20 rule” – 80% of
effects are due to 20% of causes
y More often known in the particular guise of Zipf’s law: y Special case with ζ= 1 Æ fat tails, no moments, no CLT y States that frequency is inversely proportional to rank y Often arises in economics – size of firms and cities, stock returns … y Sufficient to establish Gabaix’s propositions, but not necessary (see Sales Herfindahl below) y Zeta (ζ) is our “Cinderella parameter” – generates Gabaix’s stories
so long as 1 ≤ ζ x ) = a( x ) 2
1/ 2
−ζ / 2
= ax
y Application of Lévy’s Theoreom: The sum does not scale as √N
as it does in the CLT, but it scales as N1/ζ since the size of largest units scales as N1/ζ
Proof of Prop. 2 (cont’d) y So we have: N
N − 2 / ζ ∑ Si2 → u i =1
where u is a Lévy distributed random variable with exponent ζ/2. • Since we have N1‐/ζ = N‐2/ζ/N‐1 we obtain:
N
1−1 / ζ
⎛ ⎜N h = ⎝
−2 /ζ
N
N
∑
−1
i =1 N
∑
i =1
S
2 i
Si
⎞ ⎟ ⎠
1/2
→
d
u 1/2 E[S ]
Interpretation of Prop. 2
u 1/2 y Setting to vζ and rearranging we arrive at: E[S ]
vζ σGDP ~ 1−1/ζ σ QED. N y Economic Interpretation
If the firm size distribution has fat tails (ζ x) = x−ζ for ζ [ א1, ∞). Assume that the volatility of a firm of size S is firm −α (12)
σ
( S ) = kS
for some α > 0 (then large firms have a smaller standard deviation that small firms).
⎛ −1 ⎞ α Define: (13) ⎟⎟ α = min⎜⎜1 / 2,1 + ζ ⎠ ⎝ '
Prop. 3 (cont’d) y GDP fluctuations have the form:
ΔYt −α ' = kN g t Yt
1
if ζ >
(14)
if ζ =
1 (15)
−α '
ΔYt N =k gt Yt ln N
such that when N → ∞, gt converges to a non‐degenerate distribution. When ζ > 1, gt converges to a Lévy stable distribution with exponent min {ζ/ (1 − α) , 2}. In particular, the volatility σ (S) of GDP growth decreases as a
power law function of GDP S:
σ
GDP
(S ) ~ S
−α '
(16)
Economic Interpretation y Under the case presented by Proposition 3, and ζ = 1, large
firms are less volatile than small firms (equation 12). y The top firms in big countries are larger (in an absolute
sense) than top firms in small countries. y As the top firms determine a country’s volatility, big
countries have less volatile GDP than small countries (equation 16).
A Model with Comovement y Up to now: idiosyncratic shocks might explain a significant
portion of aggregate fluctuations y This model: see whether idiosyncratic shocks create plausibly strong comovements between the various firms or sectors of the economy y After a shock to firm i, all the other firms adjust instantaneously, instead of over time through the input‐output matrix with a lag via the general equilibrium y Derivation: o solving the social planner’s problem of maximizing utility (separable in
consumption (=output) and labor supply o subject to capital and labor constraints, and assuming prices equal marginal costs.
A Model with Comovement (cont’d) y This gives us expressions for: o aggregate output, o TFP and sales, o used to study the aggregate‐level effects of firm‐level
proportionate productivity shocks (i.e. variables are proportional changes) Æ leads us to Prop. 4
Proposition 4 y Suppose that each firm i receives a productivity shock Âi.
Macroeconomic variables change according to:
TFP:
∧
∧ Si ∧ i A Λ = ∑ Ai =∑ Sales i GDP Y
(24)
∧ 1 Λ GDP: Y = (1 − αξ ) ∧
(25)
Firm level variables change according to: ∧
∧
∧
∧
S i = X i = β Ai + (1 − βb(1 − αξ )) Y 28) Dollar Sales: (
β = 1 /(ψ − bαλ − 1 + b) where: (33)
Economic Interpretation y TFP is entirely the sum of idiosyncratic firm‐level shocks. y The economy behaves like a 1‐factor model, with an “aggregate
shock”, the GDP shock Ŷ – this shock stems from a multitude of idiosyncratic shocks y Aggregate shock causes all firm level quantities to “comove”. y Economically, when firm i has a positive shock, it makes the
aggregate economy more productive, and affect the other firms in three different ways. y First, other firms can use more intermediary inputs produced by
firm i, hence increasing their production. y Second, firm i demands more inputs from the other firms
(equation 28), which leads their production to increase.
Economic Interpretation (cont’d) y Third, given firm i commands a large share of output, it will use
more of the inputs of the economy, which tends to reduce the other firms’ output (25) The net effect depends on the magnitudes of the elasticities. y Proposition 4 have a positive loading on the GDP factor bY , i.e.
they all comove positively with GDP.‐ useful benchmark to understand comovement of the business cycle.
Empirical Analysis y A preview of the remaining part of the section: y What and how do we measure? y The models behind the growth rate y „Granular residual” y Regression results y Concentration and firm level volatility y Data: y U.S.Compustat 1951-2001 y K=100 largest firms (except energy and oil)
Gabaix’s (or Blanchard’s) story y Imagine an economy such as: y 1 big firm – 50% output and 100 small firms – 50% output y Growth rate is given (at is common shock):
g
it
= at + ε
it
y A given year: big firms’ growth 6%, small firms’s
growth 0%, so in the economy:+3% GDP y Standard answer: aggregate shock of 3% (big firm 3%, small firms -3% idiosyncratic shocks) y Gabaix shows another explaination
What and how do we measure? y Labour productivity of the firm
sales of firm i in year t zit = ln number of employees of firm i in year t y Productivity growth rate:
g it = zit − zit −1 y Now we look for models which explain the values
Models behind growth rate y 2 way of decomposition: y 1st (where f - factor proportional to GDP growth , AitTFP of firm i):
git = ft + γ Ait y 2nd (where at – shock to all firms, aIi industry specific
shock, eit – idiosyncratic shock)
g it = at + aIi ( t ) + ε it y Thanks to these: link between Ait and at, aIi and ε it
„Granular residual” y Gabaix defines 2 „Granular residuals” to measure firms’
effect on Ait (TFP) – the 2 are highly correlated y 1st (without control for industry): K
Γt =
∑ S a les
i ,t −1
i =1
( g it − g t )
K
∑ S a les i =1
i ,t −1
y 2nd (with control for industry): K
Γ ind = t
∑ Sales
i ,t −1
i =1
( g it − g Ii ( t ) )
K
∑ Sales i =1
i ,t −1
Back to Gabaix’s story y Standard model
g it = a t + ε it y
(See slide Gabaix’s (or Blanchard’s) story)
y According to Gabaix’s model: y y y
Aggregate shock 0% (standard answer 3%) 6% of idiosyncratic shock to big firm (standard answer 3%) Granular residual = 3% y
This is the effect of the big firm’s 6% growth on the GDP
Regression results (only granular residuals) y Table 1 (granular residual) y Even lagged granular residual is used (imitation of technology for instance) y Table 2 (granular residual controlled for industry) y Conclusion: the granular residual explains 1/3 of
aggregate shocks y Robustness? Yes, it’s controlled for: y Monetary shocks y Oil shocks (see Table 3 and 4 in appendix)
Table 1 (granular residual)
ε it
Conclusion: Granular residual explains GDP growth
Table 2 (granular residual controlled for industry)
Conclusion: Granular residual STILL explains GDP growth
Results in a graph
Conclusion: Different Granular residuals explain GDP growth
Stories from USA History
‘82 Volcker recession (outlier)
‘55 Car industry
’72 Steel strike ‘54 End of Korean War (outlier)
Concentration and firm-level volatility (similar thing, in a different way) y Is large firms’ volatility even big enough to explain
business cycles? (empirical question) Yes, it is. y Caculates standard deviation: y Growth of Sales/Employees ratio, Sales and Employees (12% in general) y What is the good measure of firm size? (theoretical question) y Herfindahl index is an appropriate measure (Hulton’s theorem – no need of complicated calculations) y
y Large firms volatility =>TFP volatility
Table 5 – Actual effect of idiosyncratic firm level shocks
σ GDP = ησ firm −level hS
Conclusion: 1,4% of US GDP volatility is caused by the 12% firm level volatility.
Discussion y Gabaix acknowledges existing literature proposing micro foundations for
aggregate volatility, but argues they are both conceptually and practically less useful than his “tractable” approach which is based on observables
y He notes existing work on the role of sectoral shocks in explaining aggregate
fluctuations, and debate about their relative importance empirically
y For future research on the origin of fluctuations, he suggests: y Look at shocks to big firms, not just aggregate shocks y Apply VAR and impulse response methodologies to firm‐level shocks y Think about how large‐firm shocks affect industry dynamics y Be careful about using aggregates – better information comes from firm‐level y Explaining volatility in other macro variables (inventories, etc) at firm‐level y Gabaix’s ambition seems to be to displace the Solow Residual with the Granular
Residual, but he acknowledges that his evidence is only preliminary
Appendix: Prop. 4
Appendix Prop. 4
Appendix -Table 3
Appendix - Table 4