Radical%20Innovation%20and%20Stock%20Price%20Volatility:

1

• Radical Innovation and Stock Price Volatility
• patent citation dynamics and idiosyncratic risk
  in pharma-biotech

Mariana Mazzucato (Open University) Massimiliano
Tancioni (University of Rome) Workshop on
Finance, Innovation and Inequality London,
November 9, 2007
2
Questions arising from some old and new work
TODAY Mazzucato, M. and Tancioni, M. (2006),
Stock Price Volatility Patent Citations the
case of pharma-biotech, work in progress
BACKGROUND Mazzucato, M. and Tancioni, M.
(2006), Idiosyncratic Risk Innovation a Firm
and Industry Level Analysis, OU Discussion
Paper, 50-06. Mazzucato, M. (2003), Risk,
Variety and Volatility Innovation, Growth and
Stock Prices in Old and New Industries, Journal
of Evolutionary Economics, Vol. 13 (5)
491-512. Mazzucato, M. (2002), The PC Industry
New Economy or Early Life-Cycle, Review of
Economic Dynamics, Vol. 5 318-345. Mazzucato,
M. and W. Semmler (1999), Stock Market
Volatility and Market Share Instability during
the US Automobile Industry Life-Cycle, Journal
of Evolutionary Economics, Vol. 9 67-96.

3
Aim and motivation

• To link stock price volatility dynamics to
  innovation, using firm level innovation data.
• Very little empirical work on this.
• Provide link between industry dynamics and
  financial dynamics.
• Contribute to understanding impact of real
  activity underlying stock price bubbles (vs. more
  irrational stories).

4
Innovation Knightian uncertainty

• The starting point for any financial model is the
  uncertainty facing
• investors, and the substance of every financial
  model involves the
• impact of uncertainty on the behaviour of
  investors, and ultimately, on
• market prices. (Campbell, Lo and MacKinlay,
  1997)
• Innovation ?uncertainty about future growth (high
  hopes/failures)
• RD can lead to dry hole
• Persistence one off success or new industry
  leader
• Effect on industry structure competence
  destroying innovations ? shake-up status quo
• Idiosyncratic (albeit cumulative) and tacit
  evolution of capabilities
• Uncertainty about expected growth ? Volatility.

5
Some facts on volatility

• No increase in trend of market volatility (using
  monthly value weighted composite indices
  NYSE/AMEX/Nasdaq) between 1926-2000 (Schwert
  1989 2002).
• Peaks in late 20s, 1970s oil shock, 1987 crash.
  Annual std (monthly data) 1990-9711, 1970s
  14, 1980s 16.
• However, firm specific volatility (idiosyncratic
  risk) has increased over the last 40 years
  (Campbell et al. 2001). Doubled since 1962.
  Declining correlation between individual stocks
  and decreased explanatory power of CAPM model for
  a typical stock.
• Recent papers relate firm specific volatility to
  technological change
• in a vague sense (Campbell et al. 2000 Shiller
  2000). No data.

6
Innovation ? Stock Price Volatility
Most volatility studies dont use innovation
data (just broad assumptions around impact of
innovation on uncertainty). 1. Excess
Volatility (Shiller 1981, 1989, 2000) Animal
spirits, herd behaviour, bandwagon effects.
2. Idiosyncratic Risk (Campbell et al.
2000) Increase in idiosyncratic risk since the
1960s. Why? Effect of IT revolution on speed of
information. 3. Rational bubbles (Pastor and
Veronesi 2006) Volatility rises before
idiosyncratic risk becomes systematic risk.

7
Excess Volatility and Technological Revolutions

Shiller, R.J. (1981). Do Stock Prices Move Too
Much to be Justified by Subsequent Changes in
Dividends, AER, 71.
8
Excess Volatility and Technological Revolutions

Shiller, R.J. (1981). Do stock prices move too
much to be justified by subsequent changes in
dividends? AER 71(3) 421-36 Stock prices are
5 x more volatile than can be justified by
changes in fundamentals Efficient market model
real price expected value of discounted future
dividends V the
ex-post rational or perfect-foresight price D
the dividend stream ? real discount factor
r short (one-period) rate of discount


where
9
Idiosyncratic Risk and IT Revolution
Campbell, J.Y., Lettau, M., Malkiel, B.G., and
Yexiao, X. (2000), Have Stocks Become More
Volatile? An Empirical Exploration of
Idiosyncratic Risk, Journal of Finance,
56. ----------------------------------------------
--------------------------------------------- Use
high-frequency time series data on daily stock
returns for the general market, industries and
firms for the period 1963-1997. Using variance
decomposition of a CAPM equation, decompose
return of a typical stock into market wide
return, industry specific residual,
and firm-specific residual (sum to total
volatility). Result positive deterministic
time trend in stock return variances
for individual firms not for market and industry
returns. Why? (a) IT effects on speed of
information (b) Companies have begun to issue
stock earlier in their life cycle, when
there is more uncertainty about future profits.


10

Yet none of these studies actually use
innovation data. Just assume that volatility is a
sign of uncertainty and that this is highest
during periods of technological change.

11
Firm/Industry Innovation and Stock Price
Volatility

• Uncertainty is better studied at the micro level,
  i.e. related to the
• firms environment. Evidence that most shocks
  are
• idiosyncratic to the firm or plant (Davis and
  Haltiwanger 1992).
• Look at IR and EV over the industry life-cycle
• Mazzucato, M. (2002), The PC Industry New
  Economy or Early Life-Cycle, Review of Economic
  Dynamics, Vol. 5 318-345.

12
Industry Life-Cycle (Mazzucato 2002)

13
Quality Change Autos PCs (auto Raff
/Trajtenberg 1997 Abernathy et al. 1983 PC
Berndt /Rappaport 2000)

14
Movement of 28 Leading Auto Producers Ranked
According to Places in Production (Epstein,
1928)

15
Excess Volatility in Autos (Mazzucato
2002 2004)

16
Excess Volatility in PCs (Mazzucato
2002 2004)

17
Conclusions

• Co-evolution of industrial and financial
  volatility over the industry life-cycle.
• Volatility may look irrational but tied to real
  changes in technology.
• More volatility of stock prices in phase of
  competence-destroying innovations.

18
New work using firm level innovation data

• Test for relationship between volatility and
  innovation, using firm
• level innovation data. Test across different
  industries.
• Mazzucato, M. and Tancioni, M. (2005),
  Idiosyncratic Risk
• Innovation a Firm and Industry Level Analysis,
  OU wp, 50-06.
• Is idiosyncratic risk higher for innovative
  industries and firms
• (higher RD intensity)? Sectoral taxonomy of
  innovation (Pavitt
• 1984) ? Sectoral taxonomy of stock price
  dynamics?
• ________________________________________________
  __________
• 2. Mazzucato, M. and Tancioni, M. (2006), Stock
  Price Volatility Patent
• Citations the case of pharma-biotech, work in
  progress
• Do firms with patents with higher citation
  intensity experience more
• idiosyncratic risk?
• Relation between P/E and innovation. Implications
  for bubbles.

19
Financial Data

• Industry level data Standard and Poors Analysts
  Handbook
• quarterly stock price, dividends, earnings, RD
• 34 industries
• 1976-1999
• sectoral taxonomy (RD intensity)
• __________________________________________________
  ____________________________________________
• Firm level data Compustat
• monthly stock price, dividend, earnings, RD
• 1974-2003
• annual volatility via standard deviation of 12
  month terms
• unbalanced panel
• Biotechnology (435 firms)
• Computers (129 firms)
• Pharmaceutical (282 firms)
• Textile (78 firms)

20
RD Intensity by sector, avg 1980-1992


21
Idiosyncratic Risk
Stock return for firm i

IRVolatility of firm i (or industry j) returns
vs. market M returns SP 500)
Proxy for Idiosyncratic risk for firm i
22
Methodology

• Is idiosyncratic risk higher in innovative firms
  and industries?
• INDUSTRY LEVEL
• 1. Develop 34 bivariate VAR representations of
  the industry-level
• and market-level stock returns, and perform a
  Forecast Error
• Variance Decomposition (FEVD) analysis in order
  to capture the
• degree of idiosyncratic risk of the series.
• If growth more uncertain in innovative
  industries, expect that of
• industry level predictive error variance is
  mostly explained by
• idiosyncratic (industry) shock. Expect to find
  that the forecast error
• variance explained by the market shock (i.e.
  SP500) should be
• lower in innovative sectors and higher in less
  innovative sectors.

23

2. CAPM model. Pool the industry-level data
obtaining a balanced panel with time dimension T
(88 observations) and sectional dimension N
(34 observations). Regress the industry-level
stock returns on industry specific dummies (Fixed
Effects) and the SP500 returns. Allows test
of (a) EMM, and (b) heterogeneity in
section. Expect the variability explained by the
regression to be higher for the low innovative
industries and lower for the high
innovative industries.

24

b. FIRM LEVEL Employ panel of 822 firms
belonging to 5 industries- 1974-2003 and
directly test the existence of a positive
relationship between idiosyncratic risk and
innovative effort (RD intensity). Estimate
panel regressions in which firm-level IR depends
on RD effort and the firms relative weight in
terms of market capitalization. Analysis
conducted both employing the whole panel sample
and the five different industry-specific panels
of firms.

25
Part 1 Conclusion
Industry level analysis mixed results (similar
to Campbell et al.). Expectations hold only for
some industries in extremes of the taxonomy (e.g.
very innovative semiconductors, very low
innovative public utilities). Firm level
analysis strong relationship between
idiosyncratic risk and RD intensity. Most
interesting result Relationship not stronger for
firms in more innovative industries. Relationship
holds stronger in textiles (low-innovative) than
in pharmaceuticals (highly innovative).
Perhaps because the low average RD intensity
in textiles makes innovative firms in that
industry stick out. And holds stronger in
biotech due to higher uncertainty than in
computers and pharma. Dynamic nature of
volatility computers (1989-1997) and
biotechnology (1995-2003).


26
Part 2 Patent Citation Data and Stock Return
Volatility

• Mazzucato, M. and Tancioni, M. (2006), Stock
  Price Volatility Patent
• Citations the case of pharma-biotech, work in
  progress
• Citation weighted patents measure importance of
  innovation.
• HJT question how well do patents measure
  economic performance?
• Our question are firms with higher RD
  intensity, more patents, and more important
  patents characterized by more uncertainty, and
  hence higher volatility?
• If so, this provides some insights into the real
  aspects of volatility dynamics (rather than
  animal spirits, irrational exuberance).
• Patents as signals of innovations

27
Why focus on pharma-biotech?

• Industry life-cycle approach
• A sector with high RD and high patenting rates
• Changing knowledge regimes (Gambardella 1995)
• Do volatility dynamics evolve over industry
  life-cycle with changing knowledge regimes?
• Do stock prices react to innovation more in
  random or guided search regime?

28
Data

• GIC codes 352010 for Biotech and 352020 for
  Pharmaceuticals.
• SP Compustat
• monthly stock price, dividend, earnings, RD
• compute annual volatility via standard deviation
  of 12 month terms
• unbalanced panel biotech (563 firms) and pharma
  (323 firms)
• NBER patent citation data
• Detailed patent related information on 3 million
  US patents granted between January 1963 and
  December 1999, and all citations made to these
  patents between 1975 and 1999 (over 16 million).
  
• Annual data, using application date (more
  uncertainty!)
• Merged sample 126 pharma firms and 177 biotech
  firms.

29
Variables (all in logs)

Idiosyncratic (IDRISK) stdev RET firm i /
stdev RET market Price earnings ratio (PE) RD
intensity (RDREV) RD/Revenues (both
real) Patent count (PAT) annual number of
patents for firm i divided by average number of
patents per firm in industry j. Weighted patents
(PATW) number of citations received by firm i
divided by number of patents for firm i, all
divided by same ratio for industry (avg for
firm) Patent yield (PATY) Patents/RD SIZE
Control for firm size (market share) CAPSHARE
Control for firms share of market
capitalization




30
Hypotheses (plus all the controls and dummies)

• 1. Market value is related to innovation
  indicators (RD, Patents). (Trajtenberg 1990
  Hall, Jaffe, Trajtenberg 2001)
• 2. Idiosyncratic risk is related to innovation
  indicators. (Campbell et al. 2001 Mazzucato and
  Tancioni 2005)

31
Hypotheses

• 3. Bubbles Price-Earnings related to
  Idiosyncratic Risk (Pastor and Veronesi 2004)
• 4. P/E related to innovation indicators.

32
Periods and samples tested
Periods a) Full dates 1975-1999 b)
Truncation re-run the estimates fixing the end
date 1995 (find no difference). c) Test for
possible structural breaks (e.g. institutional
changes after Bayh-Dole) using post 1985 dummy.
(significant). Test 3 different samples 1.
Whole sample (with biotech dummy) 2. Pharma
sample 3. Biotech sample

33
Results for Model 1 (MV on innovation)
RD intensity and un/weighted patents
coefficients are both positive and statistically
significant. When add patents, RD less
significant, signaling possible correlation
between RD (-2) and patents (- 1), especially
biotech. Flows are just as important as stocks
(used in HJT). Citation weighted patents not
more significant than un-weighted. Higher lag
on RD (3) than patents (1). Firm size control
positive and significant (pharma and bio). Dummy
post 1985 positive and significant (pharma and
bio). Bets fit of all models.


34
Results for Model 2 (IR on innovation)
Coefficients on RD intensity and patents
(count and weighted) are positive and
statistically significant (but less than in Model
1) Lag on RD falls to 1, i.e. volatility reacts
quicker than levels of MV and PE (3 lags in model
1 and 4), Lag on patents also falls to 1 (only
model where same lag for RD and patents). Firm
size control negative and (as expected) and
significant Post 1985 (dummy) positive and
significant only in 2a (higher volatility post
1985 as in Campbell et al. 2000, but not when
include patents).


35
Results for Model 3 (PE on IR)

• Positive and statistically significant relation
  between P/E and IR.
• Best estimates are obtained when measure of IR is
  entered
• with 2 lags.
• Suggesting that
• volatility leads levels (supported in previous
  results)
• innovation leads volatility
• Some evidence for rational bubble hypothesis

36
Results for Model 4 (PE on innovation variables)
RD intensity and weighted patents coefficients
are both positive and statistically significant,
providing support to rational bubble hyp
Better fit than Model 3. Un-weighted patents
not significant. Only model in which patent
yield is is significant! Best fit obtained
with 3 lags RD intensity, 2 lags patents.
Firm size (control) negative and significant
Dummy post 1985 positive and significant


37
Biotech differences

Biotech firms have on average 10 less
MV 30-35 more IR 5 higher P/E RD
significant only for model 4 (perhaps because
biotech is so RD intensivedont stick
out?) Patents insignificant in model 4 (except
patent yield). Slightly lower lags (market
reacts quicker to new segments of the
industry?). Stronger correlation between RD
(-2) and Patents (-1) (supported by higher mean
patent yield in biotech). Post 1985 lower P/E
(unlike pharma).

38
Conclusions

• Volatility of firm specific returns related to
  firm innovation.
• MV and PE levels also related to firm innovation.
  
• Lags IDRISK reacts more quickly to innovation
  than MV, PE quicker in Biotech than in pharma
• quicker to patents than to RD
• Dynamic correlations between RD (-2) and
  Patents (-1)
• Smaller firms have lower market value, but higher
  volatility and PE.
• Possibility of structural break post 1985
  (explore further)

39
Future extensions

• Innovation characteristics general vs. original
  
• Temporal dimension recent vs. old citations
• Does volatility react to patents more so in
  periods of high or low tech opportunity?