extending the volatilityconcept to point processes david r. brillinger statistics department
DESCRIPTION
Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department University of California Berkeley, CA [email protected] www.stat.berkeley.edu/~brill 2 π 1 ISI2007Lisbon. Introduction - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/1.jpg)
![Page 2: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/2.jpg)
Extending the VolatilityConcept to Point Processes
David R. Brillinger
Statistics Department
University of California
Berkeley, CA
www.stat.berkeley.edu/~brill
2 π 1
ISI2007Lisbon
![Page 3: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/3.jpg)
![Page 4: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/4.jpg)
Introduction
Is there a useful extension of the concept of volatility to point processes?
A number of data analyses will be presented
![Page 5: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/5.jpg)
Why bother with an extension to point processes?
a) Perhaps will learn more about time series case
b) Pps are an interesting data type
c) Pps are building blocks
d) Volatility often considered risk measure for time series.
27 July Guardian.
“Down nearly 60 points at one stage, the FTSE recovered and put on the same amount again. But by the close it had slipped
back, down 36.0 points.”
“inject billions into the banking system”
![Page 6: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/6.jpg)
Volatility
When is something volatile?
When values shifting/changing a lot
Vague concept
Can be formalized in various ways
There are empirical formulas as well as models
![Page 7: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/7.jpg)
Merrill Lynch.
“Volatility.
A measure of the fluctuation in the market price of the underlying security.
Mathematically, volatility is the annualized standard deviation of returns.
A - Low; B - Medium; and C - High.”
![Page 8: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/8.jpg)
Financial time series.
Pt price at time t
“Return” data, Yt = (Pt - Pt-1)/Pt-1
Empirical formula
Realized volatility
mean{[Ys – Ys-1]p | s near t}, p = 1 or 2
![Page 9: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/9.jpg)
Model based formula. GARCH
Yt = μt + σtεt, ε zero mean, unit variance, t discrete
σt2 = α0 + ∑ αi[Yt-i – μt-i]2 + ∑ βjσt-j
2 α’s, β’s > 0
Volatility σt2
For μs, σs smooth
mean{[Ys – Ys-1]2 | s near t} ~ σt2(εt - εt-1)2
![Page 10: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/10.jpg)
Crossings based.
E{[Y(t) – Y(t-h)] 2} = 2[c(0) – c(h)] ≈ -2c"(0)h2
Recognize as
2c(0) π2 [E(#{crossings of mean})]2
for stationary normal
Consider
s #{crossings of mean | near t}
as volatility measure
![Page 11: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/11.jpg)
Example. Standard & Poors 500.
Weighted average of prices of 500 large companies in US stock market
Events
Great Crash Nov 1929
Asian Flu (Black Monday) Oct 1997
Other “crashes”: 1932, 1937, 1940, 1962, 1974, 1987
![Page 12: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/12.jpg)
Zero crossings
![Page 13: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/13.jpg)
S&P 500: realized volatility, model based, crossing based
Tsay (2002)
n= 792
![Page 14: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/14.jpg)
Point process case.
locations along line: τ1 < τ2 < τ3 < τ4 < …
N(t) = #{τj t}
Intervals/interarrivals Xj = τj+1 - τj
Stochastic point process.
Probabilities defined
Characteristics: rate, autointensity, covariance density, conditional intensity, …
E.g. Poisson, doubly stochastic Poisson
![Page 15: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/15.jpg)
0-1 valued time series.
Zt = 0 or 1
Realized volatility
ave{ [Zs – Zs-1]2 | s near t }
Connection to zero crossings.
Zt= sgn(Yt), {Yt} ordinary t.s.
Σ [Zs – Zs-1]2 = #{zero crossings}
![Page 16: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/16.jpg)
Connecting pp and 0-1 series. Algebra
Tj = <τj/h> <.> nearest integer, embed in 0’s
h small enough so no ties
Y(t) = N(t+h) – N(t) = tt+hdN(u)
Stationary case
E{Y(t)} = pN h
cov{Y(t+u),Y(t)} = t+ut+u+h t
t+hcov{dN(r),dN(s)}
~ pNδ{u}h + qNN(u)h2
as cov{dN(r),dN(s)} = [pN δ(r-s) + qNN(r-s)]dr ds,
rate, pN, covariance density, qNN(∙), Dirac delta, δ(∙)
![Page 17: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/17.jpg)
Parametric models.
Bernoulli ARCH. Cox (1970)
Prob{Zt = 1|Ht} = πt
logit πt = Σ αi Zt-i
Ht history before t
Fitting, assessment, prediction, … via glm()
Bernoulli GARCH.
logit πt = Σ αi Zt-i + Σ βj logit πt-j
Volatility πt or πt(1 - πt)?
![Page 18: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/18.jpg)
Convento do Carmo
![Page 19: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/19.jpg)
“California” earthquakes magnitude 4, 1969-2003
N=1805
![Page 20: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/20.jpg)
Results of 0-1 analysis
![Page 21: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/21.jpg)
P.p. analysis.
Rate as estimate of volatility
Consider var{dN(t)}
var{N(t)-N(t-h)} ≈ pNh + qNN(0)h2
Estimate of rate at time t.
k(t-u)dN(u) / k(t-u)du
k(.) kernel
Variance, stationary case, k(.) narrow
pN k(t-u)2du / [ k(t-u)du]2 + qNN(0)
![Page 22: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/22.jpg)
![Page 23: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/23.jpg)
Example. Euro-USA exchange rate
![Page 24: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/24.jpg)
Interval analysis.
Xj = τj – τj-1
Also stationary
![Page 25: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/25.jpg)
California earthquakes
![Page 26: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/26.jpg)
Risk analysis. Time series case.
Assets Yt and probability p
VaR is the p-th quantile
Prob{Yt+1 – Yt VaR } = p
left tail
If approxmate distribution of Yt+1 – Yt by
Normal(0,σt)
volatility, σt, appears
Sometimes predictive model is built and fit to estimate VaR
![Page 27: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/27.jpg)
Point process case.
Pulses arriving close together can damage
Number of oscillations to break object (Ang & Tang)
Suppose all points have the same value (mark), e.g. spike train.
Consider VaR of
Prob{N(t+u) – N(t) > VaR} = 1-p
Righthand tail
![Page 28: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/28.jpg)
Examples.
S&P500: p=.05 method of moments quantile
VaR = $.0815
CA earthquakes: u = 7 days, p=.95 mom quantile
VaR = 28 events
![Page 29: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/29.jpg)
Case with seasonality – US Forest Fires 1970 - 2005
n=8481
![Page 30: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/30.jpg)
Dashed line ~ seasonal
![Page 31: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/31.jpg)
Example. Volatility, simulate Poisson, recover volatility
![Page 32: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/32.jpg)
Conclusion.
Returning to the question, “Why bother with extension?
a) Perhaps will learn more about time series case
b) Pps are an interesting data type
c) Pps are building blocks
d) Volatility often considered risk measure for time series.”
The volatility can be the basic phenomenon
![Page 33: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/33.jpg)
Another question.
“Is there a useful extension of the concept of volatility to point processes?”
The running rate
Gets at local behavior (prediction)
![Page 34: Extending the VolatilityConcept to Point Processes David R. Brillinger Statistics Department](https://reader036.vdocuments.mx/reader036/viewer/2022062301/56812d8c550346895d92a555/html5/thumbnails/34.jpg)
Some references.
Bouchard, J-P. and Potters, M. (2000). Theory of Financial Risk and Derivative Pricing. Cambridge
Dettling, M. and Buhlmann, P. “Volatility and risk estimation with linear and nonlinear methods based on high frequency data”
Tsay, R. S. (2002). Analysis of Financial Time Series. Wiley