Paper On Mcap To GDP
Paper On Mcap To GDP
Paper On Mcap To GDP
ABSTRACT
In a 2001 interview, Warren Buffett suggested that the ratio of the market value of all publicly
traded stocks to the Gross National Product could identify potential overvaluations and underval-
uations in the US equity market. In this paper, we investigate whether this ratio is a statistically
stock market crashes, equity markets, market value-to-GNP ratio, equity market, GNP, GDP, like-
lihood ratio test, Monte Carlo.
Lleo is with the Finance Department , NEOMA Business School, and Ziemba is the Alumni Professor of Financial
Modeling and Stochastic Optimization (Emeritus), University of British Columbia, Vancouver, BC and Distinguished
Visiting Scholar, Systemic Risk Centre, London School of Economics.The first author gratefully acknowledges the
support from Region Champagne Ardennes and the European Union through the RiskPerform grant.
Pointing to the spectacular rise in the level of the S&P500, which has more than tripled since
the March 9th, 2009 trough and rose nearly 40% in 2013 and 2014 alone, investors and pundits
worry that the US equity market might be dangerously overvalued. In fact, legendary investment
manager George Sorros already entered a $2.2 billion put position on the U.S. stock market as early
as August 20141 . Is the U.S. equity market about to crash?
In an attempt to answer this question, investors and pundits have turned to a variety of mea-
sures, ranging from Nobel laureate Robert Shillers Cyclically-Adjusted Price-to-Earnings (CAPE)
(Campbell and Shiller, 1988, 1998; Shiller, 2006, 2015) to Warren Buffetts ratio of the market value
of all publicly traded stocks to the current level of the GNP (MV/GNP) (Buffett and Loomis, 2001).
In this paper, we investigate whether the MV/GNP ratio is an accurate predictor of equity
market downturns, defined as a drop of more than 10% in the value of the S&P 500 within a year.
To that end, we use the likelihood ratio test proposed by Lleo and Ziemba (2015) to assess the
accuracy of the CAPE and of the Bond-Stock Earnings Yield Differential.
The paper is organised as follows. Section I presents the MV/GNP ratio and addresses a com-
mon misspecification: replacing GNP by GDP. In section II, we give a definition of equity market
downturns and list the 19 downturns that occurred between the last quarter of 1970 and the first
quarter of 2015. In Section III, we transform the MV/GNP ratio into six testable downturn pre-
diction models. Section IV presents the main statistical test step by step, from the construction
of the predictive sequence to the maximum likelihood estimate of the models accuracy, and to the
likelihood ratio test. Finally, we perform a Monte Carlo study to address the small sample bias in
Section V.
The ratio has certain limitations in telling you what you need to know. Still, it is
probably the best single measure of where valuations stand at any given moment. And
as you can see, nearly two years ago the ratio rose to an unprecedented level. That
should have been a very strong warning signal.
In this article, Buffett and Loomis (2001) follow up on an earlier interview (Buffett and Loomis,
1999), discussing the Dot.Com bubble and stock (over)valuation.
2
The idea behind the MV/GNP ratio is to gauge the total market value of companies against
the value of the goods and services that these companies produce. The market value of all pub-
licly traded US securities reflects the capacity of US firms to generate revenue, and translate these
revenues into stable earnings. The US GNP represents the market value of all the products and ser-
vices produced by US citizens and companies regardless of where they are produced. By contrast,
the US GDP is the market value of all the products and services produced in the US, regardless of
who produced it. To illustrate, the production of Apple in China would be part of the US GNP
but not GDP, while the cars produced in the United States by Toyota would count in the US GDP
but not GNP. This argument justifies the use of the GNP in the ratio.
How different would the ratio be if one used GDP instead of GNP? Figure 1 shows two compar-
ison of the US GDP and GNP between the last quarter of 1970 and the first quarter of 2015. Panel
(a) shows the evolution of the GDP and GNP on a quarterly and seasonally-adjusted basis. The
difference between the two measures is small, with GNP rising slightly above the GDP in recent
years. Panel (b) shows that the GNP-to-GDP ratio has remained in a narrow 1.00 to 1.02 range
over the whole period. This suggests that using the US GDP rather than the GNP does not have
a material impact on the results.
We confirm this intuition with an ordinary least square regression of the quarterly GNP against
the quarterly GDP. The regressions R2 coefficient is 0.9999 and the F-statistic is 3,086,864 with
176 degrees of freedom, indicating a very close fit. The slope of the regression line is 1.011, with a
standard error of 0.00058. Although it is significantly different from 1 at a 99% confidence level, it
is not significantly different from 1.011.
If the percentage relationship falls to the 70% or 80% area, buying stocks is likely to
work very well for you. If the ratio approaches 200%as it did in 1999 and a part of
2000you are playing with fire. As you can see, the ratio was recently 133%.
Figure 2 shows the ratio of the Wilshire 5000 Full Cap Price Index, a proxy for the total market
value of U.S. stocks, to the US GNP. The ratio reached its peak at 142% in March, 2000. The
figure also indicates the 80% and 120% levels, which are often used by partitioners and pundits.
The ratio has only risen above 120% on two instances over the past 45 years: during the Dot.Com
bubble and since the end of 2014. Market commenters and investment managers have taken this
pattern as evidence that the stock market is heading for a sharp decline.
3
We investigate these claims in the next sections. In our analysis, we use quarterly, seasonally-
adjusted, final GNP data from the fourth quarter of 1970 up to the first quarter of 2015, for a total
of 177 quarters. We use the Wilshire 5000 Full Cap Price Index as a proxy for the market value
of all publicly traded securities. Both datasets come from the FRED economic database at the
Federal Reserve of Saint Louis.
Using financial market and macroeconomic data series in the same measure creates a problem
of syncronicity. Financial market data, such as Wilshire 5000, are directly observable and readily
available. Macroeconomic data series, such as the GDP or GNP, are released with a lag and subject
to revisions. Hence, we cannot determine the exact value of the MV/GNP ratio using only infor-
mation available on June 30th. With this caveat in mind, it is still best to use both the June 30th
market value of the Wilshire 5000 and the final release of Q2 GNP for the purpose of testing the
accuracy of the MV/GNP ratio as a predictor of market downturn. The reason is that both dataset
include the most accurate measurement of market capitalisation and economic activity. Intuitively,
if the measure does not perform well with full information, then we cannot expect it to perform
well with partial information.
To determine the timing of a equity market corrections, we use the daily level of the S&P 500
Total Return Index from January 1st, 1970 until March 31, 2015. We obtained this data from
Bloomberg and cross-referenced it with Thomson DataStream.
4
the 19 downturns that occurred between October 1, 1970 and March 31, 2015. On average, a
downturn lasted 197 days, causing a 20.9% decline in the value of the S&P 500.
where K(t) is a time-varying threshold for the signal. Three key parameters define the signal:
The logarithm has several advantages. First, it converts products into sums and ratios into
subtractions. Second, the logarithm rescales large values into smaller ones. Third, the logarithm
of the market value and the logarithm of the GNP have an economic interpretation, respectively
as the log return on the market value of all publicly traded securities, and as the log growth of
the GNP. This means that we can interpret logMV/GNP as the difference in log return between a
financial and a productive investment. Clearly, financial returns cannot outpace yields on produc-
tive investments for extensive periods of time.
In both cases, we compute the measure based on end-of period values. Our study is concerned
with large deviations. Taking average values limits the amplitude and number of signals.
5
B. The Threshold K
Buffett and Loomis (2001) do not indicate a firm threshold on the upside. We use a threshold
of 120% in our study. This level is often used by practitioners, probably because of its apparent
symmetry with Buffetts 70% to 80% downside rule.
We also test the MV/GDP measure using two time-varying thresholds. Following earlier works
by Ziemba and Schwartz (1991) and Lleo and Ziemba (2012, 2015), we define the threshold as a
confidence level, either using a standard 95% one-tail confidence interval based on a Normal distri-
bution, or using Cantellis inequality, a one-tailed version of Chebyshevs inequality (see Grimmett
and Stirzaker, 2001, Problem 7.11.9).
Cantellis inequality relates the probability that the distance between a random variable X and
its mean exceeds a number k > 0 of standard deviations to provide a robust confidence interval:
1
P [X k] .
1 + k2
1
Setting = 1+k2
yields
" r #
1
P X 1 .
The parameter provides an upper bound for a one-tailed confidence level on any distribution.
We compute the sample mean and standard deviation of the distribution of the measures as
a moving average and a rolling horizon standard deviation respectively, as discussed by Lleo and
Ziemba (2012, 2015). Using rolling horizon means and standard deviations has the advantage
of providing data consistency. In particular, rolling horizon mean and standard deviation are not
overly sensitive to the starting date of the calculation. Most importantly, this construction addresses
the in-sample versus out-sample problem by only using past data and predetermined parameters.
Therefore, the h-quarter moving average at time t, denoted by ht , and the corresponding rolling
horizon standard deviation th are
h1
1X
ht = xti ,
h
i=0
v
u h1
h
u 1 X
t = t (xti ht )2 .
h1
i=0
In our analysis, the horizon for the rolling statistics is h = 4 quarters. We select = 25%
6
which produces a slightly higher threshold than the standard confidence interval. In a Normal
distribution, we expect 5% of the observations to lie in the right tail, whereas Cantellis inequality
implies that the percentage of outliers in a distribution may reach 25%.
C. The Horizon H
The last parameter we need to specify is the horizon H. We set the horizon to 4 quarters prior to
the local peak that preceded the equity market downturn. This approach differs slightly from Lleo
and Ziemba (2015) who set the horizon with respect to the crash identification date. The choice
reflects a tradeoff. Using the peak date as a reference point gives us a greater control over the time
horizon because we do not need to take the time between the peak and the cars identification into
consideration. However, the peak date is only a by-product of the crash identification date. As
such it is not a key determinant of the model. Overall, both choices are equivalent and yield similar
results.
The next step is to define a signal indicator sequence S = {St , t = 1, . . . , T }. This sequence
records the first quarter in a series of positive signals. The signal indicator St takes the value 1 if
the measure crosses the threshold on quarter t but not on quarter t 1, and 0 otherwise. Thus, the
event a signal occurred for there first time on quarter t is represented as {St = 1}. We express
the signal indicator sequence as the vector s = (S1 , . . . , St , . . . , ST )
7
value 1 when an equity market correction occurs and 0 otherwise. From a probabilistic perspective,
these components are Bernoulli random variables.
We denote by Ct,H the indicator function returning 1 if the crash identification date of at least
one equity market correction occurs between time t and time t + H. The relation between Ct,H
and Ct is
H
Y
Ct,H := 1 (1 Ct+i ) .
i=1
We identify the vector CH with the sequence CH := {Ct,H , t = 1, . . . , T H} and define the vector
cH := (C1,H , . . . , Ct,H , . . . CT H,H ). The number of correct predictions n is defined as
T
X
n= Ct,H = 10 cH .
t=1
The accuracy of the crash prediction model is the conditional probability P (Ct,H = 1|St = 1) of
a crash being identified between time t and time t + H, given that we observed a signal at time t.
The higher the probability, the more accurate the model. We use maximum likelihood to estimate
this probability and to test whether it is significantly higher than a random guess. We can obtain a
simple analytical solution because the conditional random variable {Ct,H = 1|St = 1} is a Bernoulli
trial with probability p = P (Ct,H = 1|St = 1).
To estimate the probability p, we change the indexing to consider only events along the sequence
{St |St = 1, t = 1, . . . T } and denote by X := {Xi , i = 1, . . . , N } the hit sequence where xi = 1 if
the ith signal is followed by a crash and 0 otherwise. Here N denotes the total number of signals,
that is
T
X
N= S t = 10 s
t=1
where 1 is a vector with all entries set to 1 and v 0 denotes the transpose of vector v. The sequence
X can be expressed in vector notation as x = (X1 , X2 , . . . , XN ). The probability p is the ratio n/N .
Table II presents the total number of signals, number of correct and incorrect predictions, and
proportion of correct and incorrect predictions for each model. The two models based on a fixed
threshold at 120% generated only two signals. The MV/GNP models based on confidence interval
and Cantellis inequality produced 11 signals each, while the logMV/GNP models based on con-
fidence interval and Cantellis inequality generated 10 signals each. These numbers are markedly
lower than the total number of equity market downturns recorded over the period. The GNP is
only released quarterly, limiting the frequency of calculation and reducing the number of signals.
8
The accuracy of the model, defined as the proportion of correct predictions, ranges from 50% for
MV/GNP with a fixed threshold to 100% for logMV/GNP with a fixed threshold. The accuracy
of the other four models is around 70%.
N
Y
L(p|X) := pXi (1 p)1Xi
i=1
N N
!
X X
L(p|X) := ln L(p|X) = Xi ln p + N Xi ln(1 p)
i=1 i=1
so the maximum likelihood estimate of the probability p = P (Ct,H |St ) is in fact the historical
proportion of correct predictions out of all observations.
We apply a likelihood ratio test to test the null hypothesis H0 : p = p0 against the alternative
hypothesis HA : p 6= p0 . The null hypothesis reflects the idea that the probability of a random,
uninformed, signal correctly predicting crashes is p0 . A significant departure above this level indi-
cates that the measure we are considering contains some information about future equity market
corrections.
9
The probability p0 is the probability to identify an equity market downturn within 4 quarters
of a randomly selected period. To compute p0 empirically, we tally the number of quarters that are
at most 4 quarters before a peak and divide by the total number of quarters in the sample. For the
entire period 1970 Q4 to 2015 Q1, we find that p0 = 36.47%. We can confirm this number heuris-
tically. Given that 19 distinct crashes occurred, then at most 4 19 = 76 quarters in our sample
fall within 4 quarters prior to a peak. Because there are 177 quarters in the dataset, the heuristic
76
probability is 177 = 42.94%, a bit higher than the empirical probability. The difference between
the heuristic and empirical probability is due to the fact that in reality equity market corrections
are not spread evenly through the period. In fact, corrections might occur in quick succession, as
was the case in the late 1990s when three crashes occurred within less than two years.
Table III presents the maximum likelihood estimates, likelihood ratio and test statistics for
each of the six models. The two models based on a fixed threshold are the two extreme cases.
With an accuracy of only 50%, MV/GNP is inconclusive. So far, logMV/GNP has a 100% track
record, but out of 2 predictions only. The other four models are significant at a 5% level. The
p-value of the MV/GNP models, at 1.46%, is slightly better than that of the logMV/GNP models,
at 3.15%. Overall, all the models using time-varying thresholds demonstrate an ability to predict
equity market downturns.
The Monte Carlo algorithm is as follows. Generate a large number K of paths. For each path
k = 1, . . . , K, simulate N Bernoulli random variables with probability p0 of obtaining a success.
Denote by Xk := Xik , i = 1, . . . , N the realization sequence where xki = 1 if the ith Bernoulli
variable produces a success and 0 otherwise. Next, compute the maximum likelihood estimate
10
for the probability of success given the realization sequence Xk as
PN k
i=1 Xi
p := ,
N
Once all the paths have been simulated, we use all K test statistics Yk , k = 1, . . . , K to produce
an empirical distributions for the test statistic Y .
From the empirical distribution, we obtain critical values at a 90%, 95% and 99% confidence
level against which we assess the crash prediction test statistics Y . The empirical distribution also
enables us to compute a p-value for the crash prediction test statistics. Finally, we compare the
results obtained under the empirical distribution to those derived using the asymptotic 2 distri-
bution.
Table IV reports the maximum likelihood estimate, empirical critical value at a 90%, 95% and
99% confidence, as well as the test statistics and p-value for each model. The results confirm the
analysis developed in the previous paragraph: all for models based on time-varying thresholds are
are significant at a 5% level. The two models based on fixed threshold provide the two extreme
cases. Overall, the likelihood ratio test based on the asymptotic 2 distribution has proved remark-
ably accurate.
VI. Conclusion
Our analysis shows that Warren Buffetts market value of all publicly traded securities as a
percentage of GNP (MV/GNP), and its parent the lorgarithm of the market value of all publicly
traded securities as a percentage of GNP (lnMV/GNP), can be a statistically significant predictors
of future market downturns. However, for these measures to work, we need to use time-varying
confidence-based thresholds rather than fixed thresholds.
This conclusion dispels a common myth about the MV/GNP ratio: that absolute level matters.
This myth has led market commentators and investment practitioners to suggest that the level
of the MV/GNP is the harbinger of an impeding market meltdown. After all, they argue, the
MV/GNP has only been higher than its current level once: in the wake of the Dot.Com bubble.
11
Our findings indicate that this claim cannot be substantiated. Using an arbitrary threshold fixed
at 120%, the MV/GDP would have signalled at most 2 out of the 19 equity market corrections that
occurred between 1970 Q4 and 2015 Q1. This is not to say that the absolute level of the ratio is
altogether irrelevant. However, in itself, the level does not provide sufficient evidence to forecast
most equity market downturns.
A major and very practical limitation of the MV/GNP ratio is its reliance on the GNP, which
is only released quarterly and subject to revisions. This reliance prevents more timely measure-
ments which are crucial to identify and anticipate possible market downturns. The recent launch
of the Atlanta Feds GDPNow responds to a rising interest in more frequent data releases. Until a
GNPNow is available, our work suggests that we can use GDPNow as a proxy without a significant
loss of accuracy.
It is time to go back to the worried investors and pundits that we met in the introduction. If
the MV/GNP ratio is currently sending a mixed signal, what do other measures tell us? At the
time of writing, Rober Shillers CAPE2 was at 27, much higher than its historical mean at 16.62,
but still lower than its Black Tuesday peak reached at 32.56 in September 1929, or its Dot.Com
peak reached at 44.20 in December 1999. Neither the BSEYD nor William Ziembas proprietary
put-call measures indicated that an equity market downturn was imminent. To summarise, none
of the measures was signalling an impeding market meltdown.
This investigation has also uncovered four additional questions. First, if the MV/GNP is a sta-
tistically significant predictor of equity market downturns, can we also use it to predict abnormally
high returns? Second, we have established that the MV/GNP ratio works in the U.S. market. Does
it also work in international markets? Third, what would be the loss of accuracy if instead of using
the final GNP release we used a consensus forecast or a forecasting model? Fourth and finally,
would combining several measures produce a more accurate forecast of equity market downturns?
12
REFERENCES
Buffett, W., and C. Loomis, 1999, Warren Buffett on the stock market, FORTUNE Magazine .
Buffett, W., and C. Loomis, 2001, Warren Buffett on the stock market, FORTUNE Magazine .
Campbell, J. Y., and R. J. Shiller, 1988, Stock prices, earnings, and expected dividends, The Journal
of Finance 43, 661676, Papers and Proceedings of the Forty-Seventh Annual Meeting of the
Campbell, J. Y., and R. J. Shiller, 1998, Valuation ratios and the long-run stock market outlook,
Grimmett, G., and D. Stirzaker, 2001, Probability and Random Processes (Oxford University Press).
Lleo, S., and W. T. Ziemba, 2012, Stock market crashes in 20072009: were we able to predict
Lleo, S., and W.T. Ziemba, 2015, Does the bond-stock earnings yield differential model predict
equity market corrections better than high p/e models?, Technical report.
Shiller, R. J., 2006, Irrational exuberance revisited, CFA Institute Conference Proceedings Quarterly
23, 1625.
Shiller, R. J., 2015, Irrational Exuberance, third edition (Princeton University Press).
Shiryaev, A. N., M. Zitlukhin, and W. T. Ziemba, 2014, When to sell Apple and the Nasdaq 100?
trading bubbles with a stochastic disorder model, Journal of Portfolio Management 40, 110.
Shiryaev, A. N., M. Zitlukhin, and W. T. Ziemba, 2015, Land and stock bubbles, crashes and exit
Ziemba, W.T., and S.L. Schwartz, 1991, Invest Japan (Probus, Chicago).
13
Notes
1 http://www.cnbc.com/2014/08/15/george-soros-loads-up-on-bearish-market-bet.html
14
1.00
1.02
0.99
1.01
1970
US
GDP
and
GNP
In
billions
of
USD
1971
1972
1973
$
2,000.00
$
4,000.00
$
6,000.00
$
8,000.00
$
10,000.00
$
12,000.00
$
14,000.00
$
16,000.00
$
18,000.00
$
20,000.00
$
0.00
1974
1975
1970
1971
1976
1972
1977
1973
1978
1974
1979
1975
1980
1976
1981
1977
1981
1978
1979
1982
1980
1983
1981
1984
1982
1985
1983
1986
1984
1987
1985
1986
1988
1987
1989
1988
1990
1989
1991
1990
1992
1991
1992
1992
1993
15
1993
1994
1994
1995
1995
1996
1996
1997
1997
1998
1998
1999
1999
2000
2001
2000
2002
2001
2003
2002
2004
2003
2005
2003
2006
2007
(a) US GDP and GNP (1970 Q4-2015 Q1)
2004
2005
2008
2009
2006
2010
2007
GNP
2014
2014
Figure 1. Comparison of the US GDP and GNP during the period 1970 Q4 to 2015 Q1 period.
Wilshire
5000
-to-GDP
ra2o
0%
20%
40%
60%
80%
100%
120%
140%
160%
1971
1972
1973
1974
1975
1977
1978
1979
1980
1981
1982
1984
1985
1986
1987
1988
1989
1991
1992
1993
1994
1995
1996
16
1998
1999
2000
2001
2002
2003
2005
2006
2007
2008
2009
2010
2012
2013
2014
120%
level
Wilshire-to-GNP
Crash Peak Date S&P Index at Trough date S&P Level at Peak-to-trough Peak-to-trough
Identification Peak trough decline (%) duration
Date (in days)
1 1971-08-04 1971-04-28 104.77 1971-11-23 90.16 13.9% 209
2 1973-04-27 1973-01-11 120.24 1974-10-03 62.28 48.2% 630
3 1975-08-08 1975-07-15 95.61 1975-09-16 82.09 14.1% 63
4 1977-05-25 1976-09-21 107.83 1978-03-06 86.9 19.4% 531
5 1978-10-26 1978-09-12 106.99 1978-11-14 92.49 13.6% 63
6 1979-10-25 1979-10-05 111.27 1979-11-07 99.87 10.2% 33
7 1980-03-10 1980-02-13 118.44 1980-03-27 98.22 17.1% 43
8 1981-08-24 1980-11-28 140.52 1981-09-25 112.77 19.7% 301
9 1984-02-13 1983-10-10 172.65 1984-07-24 147.82 14.4% 288
17
10 1987-10-15 1987-08-25 336.77 1987-12-04 223.92 33.5% 101
11 1990-01-30 1989-10-09 359.8 1990-01-30 322.98 10.2% 113
12 1990-08-17 1990-07-16 368.95 1990-10-11 295.46 19.9% 87
13 1997-10-27 1997-10-07 983.12 1997-10-27 876.99 10.8% 20
14 1998-08-14 1998-07-17 1186.75 1998-08-31 957.28 19.3% 45
15 1999-09-29 1999-07-16 1418.78 1999-10-15 1247.41 12.1% 91
16 2000-04-14 2000-03-24 1527.46 2001-04-04 1103.25 27.8% 376
17 2007-11-26 2007-10-09 1565.15 2009-03-09 676.53 56.8% 517
18 2010-05-20 2010-04-23 1217.28 2010-07-02 1022.58 16.0% 70
19 2011-08-04 2011-04-29 1363.61 2011-10-03 1099.23 19.4% 157
Table I The S&P 500 Index experienced 19 crashes between October 1, 1970 and March 31, 2015.
Model Total number of signals Number of correct pre- Proportion of correct Number of incorrect Proportion of incorrect
dictions predictions (%) predictions predictions (%)
18
(confidence inter-
val)
logMV/GNP 10 7 70.00% 3 30.00%
(Cantelli)
Table II Proportion of Correct and Incorrect Predictions for Each Signal Model The Total Number of Signal reported in Column
2 is a tally of the distinct signals. It is calculated as the sum of all the entries of the indicator sequence S. The Number of Correct Predictions in
Column 3 is the tally of signals that preceded a crash. It is calculated as the sum of all the entries of the indicator sequence X. The Proportion
of Correct Prediction in Column 4 measures the accuracy of the signal. It is computed as the ratio of column (3) over column (2). The Number of
Incorrect Predictions reported in Column 5 counts the number of signals that did not predict a crash. It is calculated as column (2) minus column
(3). Finally, the Proportion of Incorrect Predictions in Column 6 measures the percentage of false positive given by the signal. It is computed as the
ratio of column (5) over column (2).
Model Total number Number of ML Estimate p L(p) Likelihood Test statistics p-value
of signals correct ratio 2 ln
predictions
MV/GNP (fixed threshold) 2 1 50.00% 0.25 0.9268 0.1521 69.66%
MV/GNP (confidence inter- 11 8 72.73% 1.59E-03 0.0505 5.9697** 1.46%
val)
MV/GNP (Cantelli) 11 8 72.73% 1.59E-03 0.0505 5.9697** 1.46%
logMV/GNP (fixed thresh- 2 2 100.00% - - - -
old)
logMV/GNP (confidence in- 10 7 70.00% 2.22E-03 0.0990 4.6260** 3.15%
terval)
logMV/GNP (Cantelli) 10 7 70.00% 2.22E-03 0.0990 4.6260** 3.15%
* significant at the 10% level;
19
** significant at the 5% level;
*** significant at the 1% level.
Table III Maximum Likelihood Estimate and Likelihood Ratio Test: Uninformed Prior. The Total Number of Signal is
calculated as the sum of all the entries of the indicator sequence S. The Number of Correct Predictions is the tally of crashes preceded by the signal.
It is calculated as the sum of all the entries of the indicator sequence X. The Maximum Likelihood estimate p is the probability of correctly predicting
a crash that maximises the likelihood function of the model. It is equal to the ratio of the number of correct prediction to the total number of signals.
L(p0 |X)
L(p) is the likelihood of the crash prediction model, computed using the maximum likelihood estimate p. The likelihood ratio = L(p= p|X) is the
ratio of the likelihood under the null hypothesis p = p0 to the likelihood using the estimated probability p. The estimated test statistics, equal to
2 ln , is asymptotically 2 -distributed with 1 degree of freedom. The p-value is the probability of obtaining a test statistic higher than the one
actually observed, assuming that the null hypothesis is true. The degree of significance and the p-value indicated in the table are both based on this
distribution. The critical values at the 95%, 99% and 99.5% level are respectively 3.84, 6.63 and 7.88.
Signal Model Total number ML Es- Critical Value: Critical Value: Critical Value: Test statistics Empirical
of signals timate 90% confi- 95% confi- 99% confi- 2 ln (p0 ) p-value
p dence dence dence
MV/GNP (fixed threshold) 2 50.00% 4.0347 4.0347 4.0347 0.1521 53.64%
MV/GNP (confidence inter- 11 72.73% 3.3300 4.3887 9.5396 5.9697** 2.30%
val)
MV/GNP (Cantelli) 11 72.73% 3.3300 4.3887 9.5396 5.9697** 2.30%
logMV/GNP (fixed thresh- 2 100.00% 3.9077 3.9077 3.9077 - -
old)
logMV/GNP (confidence in- 10 70.00% 3.6817 3.6817 9.0733 4.6260** 3.88%
terval)
logMV/GNP (Cantelli) 10 70.00% 3.6817 3.6817 9.0733 4.6260** 4.20%
* significant at the 10% level;
20
** significant at the 5% level;
*** significant at the 1% level.
Table IV Monte Carlo Likelihood Ratio Test. The Total Number of Signal is calculated as the sum of all the entries of the indicator
sequence S. The Maximum Likelihood estimate p is the probability of correctly predicting a crash that maximises the likelihood function of the
model. It is equal to the ratio of the number of correct prediction to the total number of signals. Columns 4 to 6 report the critical values at a 90%,
95% and 99% confidence level for the empirical distribution generated using K = 10, 000 Monte-Carlo simulation. The test statistics in column 7
L(p0 |X) 1
L( 12 |X)
is equal to 2 ln (p0 ) = 2 ln L(p= p|X) and that in column 9 is 2 ln 2 = 2 ln L(p=p|X) . The level of significance indicated for both tests are
based on the empirical distribution. The p-value is the probability of obtaining a test statistic higher than the one actually observed, assuming that
the null hypothesis is true. The degree of significance indicated in the test statistics column and the p-value indicated in the table are both based on
and empirical distribution generated through Monte-Carlo simulations.