Big Data and Machine Learning QWAFAFEW September 22 2016 Blair Hull 1 Robert Merton 1997 Nobel Prize in Economics In 1980 calls attempts to estimate the equity premium a fools errand ID: 528445
Download Presentation The PPT/PDF document "Market Timing," is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Market Timing, Big Data and Machine Learning
QWAFAFEWSeptember 22, 2016Blair Hull
1Slide2
Robert Merton, 1997 Nobel Prize in Economics In 1980, calls attempts to estimate the equity premium a “fools errand”
Market Timing2Slide3
Paul Samuelson, 1970 Nobel Prize in EconomicsSaid in 1994, “Participation in market timing implies a degree of self-confidence bordering on hubris and self-deception”
3Slide4
Burton Malkiel, author of A Random Walk Down Wall Street
Said in 2013, “Don’t try to time the market. No one can do it. It’s dangerous.”4Slide5
FIRST TRADING EXPERIENCE
$500 INVESTMENT
5Slide6
STRATEGY
6Slide7
THE NUMBERS: 5 Years
50
d
ays/year
100 hands/hour
250,000 hands
Advantage
.80
Sharpe Ratio 6
7Slide8
SECOND TRADING EXPERIENCE
8Slide9
Edward O. Thorp & Neil Block – Fear Versus Greed in the Stock MarketBlair Hull – Stock market Timing & Gambling
1981- 5TH NATIONAL CONFERENCE ON GAMBLING & RISK TAKING9Slide10
There is significant potential to time the market but it is unlikely the risk adjusted returns will compete with the returns of blackjack or options market making.FINDINGS
10Slide11
250 Employees26 Exchanges9 Countries30,000 transaction/day
Filed S-1 to go publicHULL TRADING – JUNE, 199911Slide12
NYT - July 13, 1999
Goldman Sachs Group Inc. signaled its support yesterday for new ways of trading securities when it announced that it would buy Hull Group Inc., a leading electronic trading company, for $531 million.
12Slide13
13Slide14
Nobel Prize Winners:
Can they be wrong?
14Slide15
Data Explosion
Predictive AnalyticsEvolution of Academic LiteratureWhat has changed:
15Slide16
“Data Deluge”
16Slide17
“Every part of your business will change based on what I consider predictive analytics of the future.”
Genni RomettyMachine Learning17Slide18
“Predictive Policing tries to stop violent crime before it happens.”
Business Insider 09/25/201518Slide19
A Practitioner’s Defense of Return PredictabilityMay 30, 2015
By: Blair Hull – Hull Investments, LLC Xiao Qiao – University of Chicago Booth School of BusinessSSRN Link: http
://ssrn.com/abstract=2609814
Our Contribution
19Slide20
It is possible to time the market, & beneficial to do so.Double the return with half the risk
A Practitioner’s Defense of Return Predictability:20Slide21
20 variables Select variables according to correlation screen (.10)
Build regression model every 20 daysProcedure: Walk-Forward Simulation
21Slide22
Bulk of data from Bloomberg, Federal Reserve Bank of St. Louis, U.S. Census BureauShort interest of Rapach
, Ringgenberg, and Zhou (2015) from Matt RinggenbergConstruct 20 variables from the predictability literaturePrice ratios: dividend yield, price to earnings, CAPE, etcRates: bond yield, default spread, term spread, etc
Real economy: Baltic Dry Index, new orders/sales, cay
Technical: moving average, PCA-tech
Sell in May, variance risk premium, CPI, short interest
22
DataSlide23
We use daily, weekly, monthly and quarterly dataOverlapping data
Trade everyday on the auctionReplication23What is Different in this PaperSlide24
Wealth Accumulation and Positions of the Correlation Screening Model
24Slide25
Performance of
Market-Timing Strategies, 6/8/2001-5/4/2015
CS
RTCS
SPY
Return
12.11%
11.66%
5.79%
Sharpe Ratio
0.85
0.88
0.21
Max Drawdown
21.12%
21.83%
55.20%
CS = Correlation Screening Model
RTCS = Real-Time Correlation Screening Model
25Slide26
Annual Returns of Market-Timing Strategies, 6/8/2001-5/4/2015
CS
RTCS
SPY
2001
1.75%
4.45%
-8.47%
2002
3.72%
16.30%
-21.59%
2003
9.16%
-1.43%
28.19%
2004
5.91%
0.61%
10.70%
2005
2.13%
-0.22%
4.83%
2006
7.44%
4.40%
15.85%
2007
8.53%
2.85%
5.15%
2008
18.96%
23.85%
-36.69%
2009
40.32%
40.82%
26.36%
2010
2.21%
3.76%
15.06%
2011
7.69%
7.99%
1.90%
2012
15.47%
15.47%
15.99%
2013
34.79%
34.79%
32.31%
2014
14.64%
14.64%
13.47%
2015
2.45%
2.45%
2.85%
26Slide27
Two of the 3 largest drawdowns are in test periodData set too smallWere we just lucky?
27Too Good to be True?Slide28
Short Term ModelsEnsemble Methods Adaptive Systems
28GOOD NEWSSlide29
Model
Category
Horizon
Type
Weight
A
Economic/Fundamental
Long Term
Regression
80%
B
Economic
Medium Term
Weighted Regression
20%
C
Statistical
Short Term
Nonlinear Regression
15%
D
Short Term Omnibus
Short Term
Classification
5%
E
Event Based and Seasonal
Event
Mixed/Optimized Weighting
25%
F
Volatility
Short Term (Extremes)
Classification
7.50%
G
Volatility
Short Term
Regression
7.50%
H
Pure Sentiment
Short Term
KNN Regression
15%
I
Statistical
Short Term
Classification
5%
29
MODEL ENSEMBLESSlide30
“The Adaptive Market Hypothesis implies that because the risk/reward relation varies through time, a better way to achieve a consistent level of expected returns is to adapt to changing market conditions”
30ANDREW LO (2004)Slide31
Nobel Prize Winners among others say – No one can time the marketBig Data and New Technology may make it possibleAcademic literature has shifted
Summary31Slide32
Just as it was considered irresponsible to time the market in the last 30 years, it will be considered irresponsible NOT to time the market in the next 30 years.
Final Thought32