Modern’Stas+cs,’Inves+ng’and’ Wealth’louisville.edu/sphis/bb/src-2013/program/Slides_Katherine.pdf• Stas+cal’methods’generally’do’notaccount...

Modern Sta+s+cs, Inves+ng and Wealth

Katherine Benne8 Ensor, Ph.D. Director, Center for Computa+onal Finance

and Economic Systems (CoFES) Department of Sta+s+cs

Rice University

Southern Regional Council on Sta+s+cs 2013 Summer Research Conference

What a celebra+on!!!!

2013 SRCOS Summer Research Conference, K. Ensor 2

“For Today’s Graduate…” For Today’s Graduate, Just One Word: Sta$s$cs Steve Lohr, NYT, August 6, 2009

“I keep saying that the sexy job in the next 10 years will be sta+s+cians,” said Hal Varian, chief economist at Google. “And I’m not kidding.”

Yet data is merely the raw material of knowledge. “We’re rapidly entering a world where everything can be monitored and measured,” said Erik Brynjolfsson, an economist and director of MIT’s Center for Digital Business. “But the big problem is going to be the ability of humans to use, analyze and make sense of the data.”

The new breed of sta$s$cians tackle that problem. They use powerful computers and sophis+cated mathema+cal models to hunt for meaningful paMerns and insights in vast troves of data.


Modern Sta+s+cs, 2012

•  Complex data types and structures •  Massive data to “li8le” data •  Difficult dependencies in all dimensions •  Mixtures •  Dynamic and evolving •  Robust

–  Sta+s+cal Graphics –  Mul+variate dynamic +me series –  Bayesian methods –  Nonparametric methods –  Nonparametric Bayesian Methods –  Simula+on and Resampling –  Sta+s+cal / machine learning –  Network models (Neural, Bayes, General) –  Regression trees –  Hierarchical models –  Func+onal Data Analysis –  Categorical methods –  Nonlinear regression –  Dependent series –  Stochas+c processes –  Survival Analysis –  Agent based model –  Anomaly detec+on


Ensor, 2013, WIRES

What is a Data Scien+st? By Amazon’s John Rauser, Forbes, Oct. 8, 2011 •  More data is be8er… understanding what is “more” •  Training of “data scien+sts”: Curiosity, Communica+on and Skep+cism

–  “If you have a healthy skep+cism, you will look as hard for evidence that refutes your thesis as you will for evidence that confirms it,” Rauser said. There is a reason that “born skep+c” is a common expression. But are all skep+cs born, rather than made? How to acquires skep+cism

–  Can it be taught? Rauser says we’re in luck, ci+ng the applied sta+s+cal compu+ng course at Rice University, taught by one Hadley Wickham, inventor of the ggplot2 sta+s+cal visualizaiton program, based off the R sta+s+cs compu+ng language.

–  Wickham places a value on skep+cism that encourages it as a learned behavior.

•  If a project uncri+cally accepts its findings, it gets an “F.” If a project is cri+cal of its findings and uses “mul+ple approaches and techniques to verify unintui+ve results,” an “A+” is awarded.

•  And I would add rigor and reproducibility 2013 SRCOS Summer Research Conference,

K. Ensor 5

The “new breed” of sta+s+cians

•  Like the “old breed”!!! •  Understand the limita+ons and complexi+es of the methodologies and algorithms they are using

•  Understand the limita+ons of the data itself –  How was it collected? How should it be collected? What does the data truly measure? What sta+s+cal methodologies work for the specific type of data?

–  Sampling and experimental design are essen+al subjects –  Sadly an oken overlooked issue

•  Discovering false pa8erns HELPS NO ONE -‐-‐ and can have disastrous consequences


The Fall of Enron Timeline • 10/4 our methods identify the futility of the situation • 10/16 announced nonrecurring losses of $1billion • Competing state of the art statistical tools, pinpoint the problem on 11/28 after junk status is achieved

11/21 junk status 12/1 bankruptcy

An example: Iden+fying Anomalous Behavior Koev and Ensor, 2006


DID WE LEARN? MAYBE, SORT OF…


•  Recipe for Disaster: The Formula that Killed Wall Street, Wired Magazine, 2/23/2009

•  Allowed “hugely complex risks to be modeled with more ease and

accuracy than ever before”. •  “But it's a very inexact science. Just measuring those ini+al 5 percent

probabili+es involves collec+ng lots of disparate data points and subjec+ng them to all manner of sta+s+cal and error analysis.”

•  Failed to capture the true joint probability of default … strong assump+ons that were not met when market dynamics changed.

•  Copula is also used at the regulator level – and extensively by Moody’s for ra+ngs at the +me (NISS and OCC Explora+ons Workshop: Financial Risk Modeling and Banking Regula+ons, Feb. 2009)

Modern Methods Used Poorly….

P [TA < 1, Tb < 1] = �2(��1(FA(1)),��1(FB(1)), �)


Ill-‐behaved bonds

•  Can we come up with a different strategy for es+ma+ng the yield curve for a bond.

•  And if so, can we use this strategy to iden+fy bonds that are behaving differently than the norm of their group?

•  We are able to capture the firm level variability with a careful detailed sta+s+cal analysis.

•  Such a close look generally always leads to strong outcomes….


Modern Sta+s+cs is powerful

Time to Maturity (years)

Yie

ld (

%)

2

4

6

8

10

12

14

16

18

20

0 5 10 15 20

AAA (18)

0 5 10 15 20

AA (24)

0 5 10 15 20

A (91)

0 5 10 15 20

BBB (68)

Es$ma$ng the Term Structure With a Semi-‐parametric Bayesian Hierarchical Model: An Applica$on to Corporate Bonds Alejandro CRUZ-‐MARCELO, Katherine B. ENSOR, and Gary L. ROSNER, JASA 2011

• Es+mated Term Structure for each Corporate Bond in the Data Set • Method “borrows strength” from similar groups; outliers handled automa+cally • Classifica+on is semi-‐parametric and robust


Modern Sta+s+cs used well can make a big difference…

!"#$%&

'%(()*

+ , -+ -, .+ ., /+ /, 0+ 0,

12!*)#345

67!!*)#345

67!!83*9

:;;<=,>>?

12!*)#345

67!!*)#345

67!!83*9

:::<=/-?

12!*)#345

67!!*)#345

67!!83*9

::<=--@?

12!*)#345

67!!*)#345

67!!83*9

:<=/+A?

12!*)#345

67!!*)#345

67!!83*9

666<=-0,?

0 5 10 15 20 25 30

05

10

15

20

Maturity

%

Yield Curves by Credit Rating (termstrc)

Aaa

Aa

A

Baa

•  The above es+mated yield curves are based on methods proposed as late as 2004;

•  fall in to the class of non-‐linear regression strategies.

•  Undiscovered challenge was the mixture of bond behaviors below AAA ra+ngs. 2013 SRCOS Summer Research Conference,

K. Ensor 12

Great Sta$s$cians are Ques$oning and Inquisi$ve…

•  How Bright Promise in Cancer Treatment Fell Apart

•  New York Times, Gina Kolato, July 7, 2007

•  Keith Baggerly, (Rice Sta+s+cs Dept. Alumnus BA, MS and Ph.D.) and Kevin Coombes both of MDA Cancer Center

•  A significant breakthrough… –  Duke researchers developed

methodology to use a pa+ents own tumor cells to iden+fy which drugs would be effec+ve through the gene pa8erns.

•  The research was wrong. •  The pa+ent lost her ba8le with

cancer.

REPRODUCIBLE RESEARCH: Through a strong dose of sta+s+cal inves+ga+ve skills and core scien+fic values, Baggerly and Coombes discovered the mistake in the Duke research. Their journey in bringing this mistake to light is a lesson in itself. 2013 SRCOS Summer Research Conference,

K. Ensor 13

Simple is not always the solu+on… •  Many simple approaches rely on “old

style” sta+s+cal methodologies whose limita+ons are well understood by the well trained sta+s+cian but oken not by the novice prac++oner.

•  Consider a macro level problem based for global investment –  Convergence of global markets – typical

strategies did not properly account for the decreased diversifica+on.

–  Typical regression strategies lead to underes+mated VaRs -‐Using quan+le regressions that adapt with the changing correla+on structure.

•  Enterprise and Poli+cal Risk Management in Complex Systems, Ensor, Kyj and Marfin, The Journal of Energy Development, 2009

Simple VaR calcula+on led to a 50% underes+mate of the VaR, due to Poli+cal risk, of the $500M proposed project. 2013 SRCOS Summer Research Conference,

K. Ensor 14

Modern Sta+s+cs and Inves+ng •  Inves+ng on what scale?

–  Annual rebalancing to algorithmic trading to value based inves+ng? •  The “macro”, “micro” or “nano” scale?

•  “Modern portolio” theory relies on first and second moments and factors incorporated through first and second moments. It is no longer modern but is it s+ll useful?

•  And is this even less true given the market dynamics of today? –  The “nano” scale impacts the “micro” and the “macro” scale.

•  Can one overcome the basic limita+ons of modern por[olio theory with the proper implementa+on of modern sta+s+cs?

•  And understanding the poli+cal, economic and changing structures of financial markets?


That is our challenge… • A challenge addressed by leading researchers and prac+oners in sta+s+cs.


Consider the Nano scale Algorithmic trading

•  Aker quietly growing to account for about 60 percent of the seven billion shares that change hands daily on United States stock markets, the firms are trying to stave off the regulators who are proposing to curb their ac+vi+es. (NYT Oct. 16, 2011).

•  Does Algorithmic Trading Improve Liquidity? Hendersho8, Jones and Menkveld, Journal of Finance, 2011 –  “For large stocks in par+cular, AT narrows spreads, reduces adverse selec+on, and reduces trade-‐related price discovery. The findings indicate that AT improves liquidity and enhances the informa+veness of quotes.”

2013 SRCOS Summer Research Conference, K. Ensor

Bstrader.blogspot.com

17

Flash Crash, May 6, 2010

•  John Carter, Points and Figures, Flash Crash

Factbox: Aker the flash crash, changes to U.S. markets, Jonathan Spicer, Reuters, Sep. 1, 2011

On May 6, 2010, the Dow Jones industrial average plunged nearly 700 points in just minutes before rebounding, sending blue-‐chip stocks sharply lower and briefly wiping out an es$mated $1 trillion in market capitaliza$on. “Last year's "flash crash" brought to a boil a debate over stability and fairness in the U.S. equity marketplace.”


Lessons from the Flash Crash •  Marla Brill, Financial Advisor Magazine, August 2010 “The spotlight fell on exchange-‐traded funds, which were jolted more than

other securi+es by the collapse. About 210 of 980 ETFs traded that day at less than half their ul+mate closing price, according to Morningstar. Though the trades of all kinds of securi+es were canceled in instances where execu+on prices differed by at least 60% from pre-‐crash levels, ETFs represented some 70% of those cancella+ons. Such sta+s+cs prompted a number of ar+cles that made unfavorable comparisons between ETFs, which can suffer violent intraday vola+lity, and the more predictable mutual fund, whose net assets values get tallied up at the end of the day aker trading closes.”

“By the end of it, aker the dust had cleared and prices had stabilized, ETF

investors who had simply held on to their posi+ons hadnʼt suffered any more than those in other types of stocks or mutual funds.”


Hizng the Switch on the New Circuit Breaker

•  BILL ALPERT AND LISA STRYJEWSKI, Barron’s August 13, 2011

•  Just in the nick of +me, it seems, stock exchanges have expanded a key safety mechanism aimed at preven+ng swoons like the May 2010 "flash crash," when the Dow bizarrely dropped almost 1,000 points in 20 minutes, then snapped back on extraordinary volume. As of last Monday, every stock listed in the U.S. is covered by its own circuit breaker designed to pause a stock's trading if it makes a sudden large move.

Stuart Goldenberg for Barron's

• The proposed limit up-‐down vola+lity controls will trigger most frequently among the smallcap stocks, especially during market upheavals. • But even if the rules had been in place in the flash crash, the controls would have affected only about 14%, or 143 stocks, of the Russell 1000, and 7%, or 535, of all listed shares 2013 SRCOS Summer Research Conference,

K. Ensor 20

Electronic Gossip – The New Insider Trading Informa+on?

•  Twi8er power: How to Dominate Your Market by Joel Comm, NYT Bestselling Author

•  Twi8er Predicts the Stock Market, Aeron Saenz, 10/2010

•  If you want to make a killing on Wall Street, social media may be the secret to your success. Researchers at Indiana University and the University of Manchester have found that the moods expressed in Twi8er feeds can accurately predict some changes in the Dow Jones Industrial Average three or four days before they occur.

•  A self fulfilling prophecy?


Modeling the markets

•  Sta+s+cal methods generally do not account for the significant level of market microstructure.

•  Should it? Or can the underlying “structure” and perturba+ons in the financial markets be lek to the “noise” in the system.

•  The accumula+on of “noise” some+mes becomes signal…and a driver in the overall system.


Risk Intertwined: Systemic Risk Exists

Risks

Risks

Risks

Risks Investment(s) Investment(s) Investment(s) Investment(s) Investment(s) Investment(s) Investment(s) 2013 SRCOS Summer Research Conference,

K. Ensor 23

And will con+nue to exist… •  We cannot regulate away systemic risk; we can only change

its form. •  If market structures must change I am reminded of Pareto’s

chart used so well in quality control across the world – highligh+ng the most common sources of defects

•  Understanding the degree of global market dependence and systemic risk is key for all investors regardless of their investment style -‐-‐ diversifica+on is a moving target

2013 SRCOS Summer Research Conference, K. Ensor

A common observa+on is convergence of many financial instruments.

24

Example: Convergence of Hedge Fund Strategies

First Period

CADSB

EM

EMN

ED

DI

EDMSEDRA

FIA

GM

LSE

MF

MS

S.P.500

Second Period

CADSB

EM

EMN

ED

DI

EDMSEDRA

FIA

GM

LSE

MF

MS

S.P.500

Third Period

CADSB

EM

EMN

ED

DI

EDMSEDRA

FIA

GM

LSE

MF

MS

S.P.500

July 2007 – April 2011 Red – correla+on >.5 Black – nega+ve correla+on

4/04 12/00

1/01 6/07

Ensor, Marfin, Seidens+cker, Miller 2011


NYU STERN SYSTEMIC RISK RANKINGS •  A firm is systemically risky if it is likely to face a capital shortage just

when the financial sector itself is weak. •  Updated daily •  Similar to stress tests that are regularly applied to financial firms •  Uses publicly available informa+on and is quick and inexpensive to

compute •  The measure incorporates the vola+lity of the firm and its correla+on with

the market, as well as its performance in extremes (and yes they use copulas).


What is the “quan+ta+ve” investor to do?

•  Ignore the market nano-‐structure??? •  Or Meet the Challenge?

–  Pay a8en+on to issues of liquidity and transac+on costs –  Understand systemic risk –  Simplest fix -‐-‐ use robust strategies that are not impacted by this structure –  Target “stable” +me epochs to place longer term posi+ons –  Properly incorporate the essence of the dynamics into factor based models –  Rely on the law of large numbers and the central limit theorem and con+nue business as usual with

modern portolio theory –  Develop model free strategies that capture the dependence structure between stocks, e.g.

SIMUGRAM, Thompson et al –  Develop dynamic +me series strategies that fully capture the changing structure – if you can –  Move in to long posi+ons that poten+ally guard against global uncertainty

•  Most certainly rely on MODERN STATISTICAL methods, while fully understanding the limita+ons of the array of methods used, and maintain a healthy dose of skep+cism.

Photo downloaded – no a8ribu+on provided.


Some rules of thumb… •  Aker watching a decade of back-‐tes+ng of a mul+tude of equity portolio’s, what have I observed?

•  It does help to be strategic and invest the +me. •  Equal weigh+ng strongly outperforms market-‐cap-‐weigh+ng.

•  Ideally quan+ta+ve screens that get you to a pool of reasonable choices – and then fundamental analyses on these choices – if you have +me!

•  There are +mes when “cash is King” at least some! •  Thompson’s Simugram© and “max-‐median” rule are general “everyman” strategies for momentum style portolios.


Much of today’s global prosperity

•  Can be a8ributed to the intelligent use of modern sta+s+cs. – Agriculture, manufacturing –  In product development –  In high global construc+on and infrastructure –  In drug discovery and pa+ent care –  In generally healthier socie+es

•  And most certainly in the global financial industry 2013 SRCOS Summer Research Conference,

K. Ensor

www.squido.com

29

And remember the “new breed” of sta+s+cians

•  Like the “old breed”!!! •  Understand the limita+ons and complexi+es of the methodologies and algorithms they are using

•  Understand the limita+ons of the data itself –  How was it collected? How should it be collected? What does the data truly measure? What sta+s+cal methodologies work for the specific type of data?

–  Sampling and experimental design are essen+al subjects –  Sadly an oken overlooked issue

•  Discovering false pa8erns HELPS NO ONE -‐-‐ and can have disastrous consequences


Decisions are based on data…good use of sta+s+cs can guide those decisions •  In the financial industry every decision is based on some form of data,

whether measured or subjec+ve. •  Understanding the quality and source of data is impera+ve •  The R revolu+on puts modern sta+s+cs in the hands of anyone •  Appropriate use of modern sta+s+cs helps immensely toward improving all

decisions with regard to inves+ng – but modern sta+s+cs is complex and not well understood by the novice prac++oner.

•  And as in everything, a healthy dose of skep+cism is essen+al; let’s all make an A+ in Hadley’s course! •  And is we have heard, it is not just in the numbers… •  And in the words of John Tukey, so oken quoted by Rice’s own Jim Thompson

“BeMer an approximate answer to the right ques+on, than an exact answer to the wrong ques+on.”

•  And with that I … 2013 SRCOS Summer Research Conference, K. Ensor 31


Rice University Center for Computa+onal Finance and Economic Systems

AND Department of Sta+s+cs

Galveston Dunes

Do you recognize…..

Hotel Galvez

SAVE THE DATE! 2014 SRCOS SUMMER RESEARCH CONFERENCE

50TH ANNIVERSARY June 1-‐4th

Modern’Stas+cs,’Inves+ng’and’ Wealth’louisville.edu/sphis/bb/src-2013/program/Slides_Katherine.pdf• Stas+cal’methods’generally’do’notaccount...

Documents