Return predictability has been a controversial topic in finance for a long time. We show there is substantial predictive power in combining forecasting variables. We apply correlation screening to combine twenty variables that have been proposed in the return predictability literature, and demonstrate forecasting power at a six-month horizon. We illustrate the economic significance of return predictability through a simulation which takes positions in SPY proportional to the model forecast.
The simulated strategy yields annual returns more than twice that of the buy-and-hold strategy, with a Sharpe ratio four times as large. This application of big data ideas to return predictability serves to shift the sentiment associated with market timing.
Market Timing, Big Data, and Machine Learning by Xiao Qiao at QuantCon 2016
1. Market Timing, Big Data, and Machine Learning
Xiao Qiao
University of Chicago and Hull Investments LLC
2. Overview
“Big data” has invaded all aspects of our lives
Applications are only getting broader
Finance could significantly benefit from utilizing more
sophisticated predictive tools
Market-timing is a natural application of predictive analytics
I will present a simple but effective market-timing model
X. Qiao, Hull Investments Market Timing and Big Data 1
4. Origin Story
Where does the word “big data” come from?
John Mashey, Silicon Graphics (SGI) in the mid 1990s
Slide deck titled “Big Data and the Next Wave of InfraStress” in 1998
Mashey: “I wanted the simplest, shortest phrase to convey that the
boundaries of computing keep advancing”
Frank Diebold, University of Pennsylvania
“ ‘Big Data’ Dynamic Factor Models for Macroeconomic
Measurement and Forecasting” at the Eighth World Congress of the
Econometric Society in 2000
Used the word in an economics/statistics sense rather than a
computing context
X. Qiao, Hull Investments Market Timing and Big Data 2
6. Predictive Analytics Will Dominate
“Every part of your business will change based on what I consider
predictive analytics of the future”
Ginni Rometty
Chairwoman, President, and CEO of IBM
X. Qiao, Hull Investments Market Timing and Big Data 4
8. Deep Learning
Deep learning/neural net
Reinforcement learning
“Intelligent” system, not just raw computation power
X. Qiao, Hull Investments Market Timing and Big Data 6
9. Big Data is Everywhere
Finance could benefit from this data revolution
Both financial and non-financial data have exploded along with
the big data trend
Predictive tools applied to finance
Forecasting the stock market
X. Qiao, Hull Investments Market Timing and Big Data 7
11. 30-Second Primer on the Equity Premium
Equity Premium
Excess return that investing in the stock market provides over a
risk-free rate [Investopedia]
Why is it important?
Offers insight to understanding a key risk-reward trade-off in the
marketplace and investor preferences
Used in asset allocation
Individual and institutional investors
Input for cost of capital calculations, affect corporate decisions
X. Qiao, Hull Investments Market Timing and Big Data 8
12. Can We Forecast the Equity Premium?
NO...
Merton, Samuelson, Malkiel
Goyal and Welch (2008)
YES...
Fama, Shiller, Campbell
Cochrane (2008)
X. Qiao, Hull Investments Market Timing and Big Data 9
13. We Can’t Produce a Good Ex Post Estimate
Robert Merton, 1997 Nobel laureate:
Merton (1980) calls attempts to precisely estimate the equity
premium “a fool’s errand”
X. Qiao, Hull Investments Market Timing and Big Data 10
14. We Can’t Predict Where the Market is Going
Paul Samuelson, 1970 Nobel laureate:
“Participation in market timing implies a degree of self-confidence
bordering on hubris and self-deception”
X. Qiao, Hull Investments Market Timing and Big Data 11
15. Don’t Even Try
Burton Malkiel, author of A Random Walk Down Wall Street:
Warned the audience “Don’t try to time the market. No one can do
it. It’s dangerous.” at a conference in 2013
X. Qiao, Hull Investments Market Timing and Big Data 12
16. Counter Argument
Finance theory tells us the equity premium varies through time
Changing investment opportunity set, Merton (1973)
Linked to fluctuations in the macroeconomy
Influenced by investor disposition
We should be able to forecast this quantity using the appropriate
information set
Variables that capture the state of the economy
Variables that reveal investor expectations
Variables that show mispricing
X. Qiao, Hull Investments Market Timing and Big Data 13
17. Evidence that Favors Predictability
Many predictors proposed in literature: Price ratios, bond
spreads, technicals, etc
Fama and French (1988)
Campbell and Shiller (1988a, 1988b)
Cochrane (2008)
X. Qiao, Hull Investments Market Timing and Big Data 14
18. What Has Changed
Data explosion
The availability of both financial and non-financial data have
massive increased
Predictive analytics
Modern machine learning techniques, making sense of all of the
data
Evolution of the academic literature
Increasing evidence that market movements can be anticipated
X. Qiao, Hull Investments Market Timing and Big Data 15
19. What We Do
“A Practitioner’s Defense of Return Predictability” by Blair
Hull and Xiao Qiao
Leverage these recent changes to build a market-timing strategy!
It is possible to time the market, and beneficial to do so
Double the return of buy-and-hold with half the risk
What’s different in this paper?
Consider predictability from the investor’s perspective
Use daily, weekly, monthly, and quarterly data
Easy to replicate, data available upon request
X. Qiao, Hull Investments Market Timing and Big Data 16
21. Research Question
Participate in the return predictability debate:
Can we predict the equity premium well enough to form a trading
strategy?
Resolve the predictability debate from the investor’s perspective
Traditionally the question has been “Can we predict the equity
premium statistically”
Focus on economic magnitudes, not big t-stats
X. Qiao, Hull Investments Market Timing and Big Data 17
22. This Paper
Evaluate and combine 20 predictors from literature
Capture diverse information sets
Correlation screening to remove the least informative variables
Form rolling window forecasts of the equity premium
Construct market-timing strategy based on forecasts, invest in SPY
Backtest shows 12% annual returns in 2001-2015, twice as high as
buy-and-hold
Sharpe ratio of 0.85, four times that of buy-and-hold
Smaller drawdowns
Eliminate look-ahead bias by including variables as they are
discovered in the literature, get same results
X. Qiao, Hull Investments Market Timing and Big Data 18
23. Data
Bulk of data from Bloomberg, Federal Reserve Bank of St. Louis,
U.S. Census Bureau
Short interest of Rapach, Ringgenberg, and Zhou (2015) from Matt
Ringgenberg
Construct 20 variables from the predictability literature
Price ratios: dividend yield, price to earnings, CAPE, etc
Rates: 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
X. Qiao, Hull Investments Market Timing and Big Data 19
24. Information in Return Predictors
Correlation matrix of predictor variables. Positive correlations are in green and negative
correlations are in red.
X. Qiao, Hull Investments Market Timing and Big Data 20
25. Relationship with Market Returns
Correlation between predictors and
one-, three-, six-, and 12-month
future market returns. Positive
correlations are in green; negatives
are in red.
X. Qiao, Hull Investments Market Timing and Big Data 21
26. Econometric Model: Kitchen Sink
Kitchen Sink (KS): Include all predictors except price ratios, which are replaced by
PCA-price for a total of 16 variables
Re
m,t→t+130 = αKS + βKS xt + KS,t→t+130
xt =
x1,t
x2,t
...
x16,t
=
PCA − pricet
BYt
...
SIt
Simulated Strategy:
Fit model every 20 days to get ˆβKS
Use ˆβKS along with updated predictors to construct daily forecast
Take position in SPY proportional to forecast, adjust daily
Process repeats every 20 days
X. Qiao, Hull Investments Market Timing and Big Data 22
27. Econometric Model: Correlation Screening
Correlation Screening (CS): Include a predictor only if its correlation with 130-day
future market returns exceed 10%
Re
m,t→t+130 = αCS + βCS ˜xt + CS,t→t+130
˜xt =
x1,t I|ρ1,m|>0.1
x2,t I|ρ2,m|>0.1
...
x16,t I|ρ16,m|>0.1
ρi,m = Corr(xi,t , Re
m,t→t+130)
Real-Time Correlation Screening (RTCS): CS, only include variables that have been
discovered
Re
m,t→t+130 = αRTCS + βRTCS ˘xt + RTCS,t→t+130
˘xt = ˜xt|xi,t has been discovered
Backtest strategies the same way as KS
X. Qiao, Hull Investments Market Timing and Big Data 23
28. Realized Returns vs. Forecasts
Vertical axis: actual returns; horizontal axis: forecast returns. The black solid line is
drawn at 45 degrees. The green dashed line is the best linear fit.
X. Qiao, Hull Investments Market Timing and Big Data 24
32. Performance Comparison
KS returns the same as buying SPY, with half of the volatility
CS and RTCS double the return on SPY, with half of the risk
Market-timing strategies all have smoother return profiles than SPY,
smaller drawdown
X. Qiao, Hull Investments Market Timing and Big Data 28
33. Annual Returns
CS and RTCS have smaller ranges compared to that of SPY
X. Qiao, Hull Investments Market Timing and Big Data 29
34. Annual Returns
Need to stay disciplined...Consecutive under-performance compared to SPY
X. Qiao, Hull Investments Market Timing and Big Data 30
35. Annual Returns
Model performed well in large downturns in 2002 and 2008
X. Qiao, Hull Investments Market Timing and Big Data 31
36. Caveats
Too good to be true?
Two of the three largest market downturns are in the test period, the
third is the Great Depression
Maybe we just got lucky with 2001-2015, what about earlier periods?
Only a true out-of-sample test will remove all doubts, time will tell
X. Qiao, Hull Investments Market Timing and Big Data 32
37. Paper Summary
Combine 20 variables using correlation screening
Market-timing strategies based on return forecasts
Outperform buy-and-hold with twice the returns and half of the
volatility
Consider more sophisticated ways to combine variables
X. Qiao, Hull Investments Market Timing and Big Data 33
40. Tactical Asset Allocation
The typical advice from financial advisers is to put 60% of one’s
portfolio into equities
Optimal if expected returns were constant
Necessarily suboptimal with time-varying returns
Investors should dynamically adjust their equity position
Depending on forward-looking expected returns
Two problems arise with adjusting equity expose
Must be able to reliably forecast returns
Constantly monitor forecast to take appropriate positions
X. Qiao, Hull Investments Market Timing and Big Data 35
41. Tactical Asset Allocation
A market-timing portfolio is a superior alternative to
buy-and-hold
Automatically adjusts the equity exposure based on return
forecasts
Investors can replace their buy-and-hold allocation with the
market-timing portfolio
Don’t have to worry about monitoring forecasts
Achieve higher returns with less risk
Get the same diversification benefits across asset classes
X. Qiao, Hull Investments Market Timing and Big Data 36
42. Broader Impact of Market Timing
Investors already market time...
BUT
They tend to buy at tops and sell at bottoms
This behavior makes prices more volatile than it should be
Destabilizing to the marketplace
X. Qiao, Hull Investments Market Timing and Big Data 37
43. Benefits of Market Timing
To use this type of strategy requires a “contrarian spirit”
There can be no emotion, fear, or greed
Result in a more stable marketplace
X. Qiao, Hull Investments Market Timing and Big Data 38
44. Summary
Big data has taken over many aspects of everyday life
Its influence is only getting bigger
Finance can benefit by riding this trend
Market timing with many variables is just the tip of the iceberg
Nobel laureates say no one can time the market
Big data and new technology make it possible
Academic literature has shifted
X. Qiao, Hull Investments Market Timing and Big Data 39
46. Final Thought
It was considered irresponsible to time the market in the last 30 years,
but it will be considered irresponsible NOT to time the market in the
next 30 years.
X. Qiao, Hull Investments Market Timing and Big Data 41