How important are the rules used to create smart beta portfolios

How Important Are The Rules Used To

Create Smart Beta Portfolios?• How much did portfolio construction rules affect

performance?

• How should investors proceed?

Presented by:

Amitabh Dugar, Ph.D.

Ralph Goldsticker, CFA

Introduction

Most Smart Beta presentations are about: “What and Why?”

Presentation addresses: “Do the rules used to construct a Smart Beta portfolio matter?”

Approach:

� Use alternative portfolio construction rules to simulate multiple 25-year return histories for Low Volatility, Fundamental Indexing and Momentum strategies.

� Compare average returns, risks, drawdowns and factor exposures.

Source: 2

Proliferation of Smart Beta offerings

Number of US Large Cap Smart Beta ETFs

Smart Beta # of ETFs

Value 39

Dividends 29

Fundamental 6

Earnings 3

Revenues 1

Low Volatility, Low Risk 12

Equal Weight, Small Cap Tilt 7

Momentum 5

Quality 3

Total 66

Source: Dugar & Goldsticker classifications, ETF.com data, as of 5/1/2015 3

� How do you choose?

� How much does it matter?

Designing a smart beta strategy

Smart beta strategies are intended to provide additional exposure to attractive characteristics.

But,• True factors behind smart betas are not

observable.• Investors also care about other portfolio

characteristics.

Managers design portfolio construction rules to:• Maximize exposure to target characteristic • Manage other characteristics and behavior• Have attractive backtests

4

Large number of ways to construct any Smart

Beta strategy

5

Choices include:• Metric and formula

• Universe: Geography and capitalization range

• Subset: All stocks, top 20%, top 50%, etc.

• Weighting: heuristic, optimized, cap, equal

• Tracking error aware: optimized, sector neutral

• Constraints, concentration, liquidity

• Reconstitution frequency

� How do you decide?

� How much does it matter?

Example: Decisions required to construct a low

volatility portfolio

6

� Which measure of volatility?• Standard deviation of total, excess, or residual returns; beta

• Estimation window: 1 year, 3 years, 5 years

� How is the portfolio constructed?• Heuristic rule-based

• All stocks weighted by inverse volatility

• Reweighted subset

• Which subset? All stocks; top 20%, 50%; sector neutral, …

• Which weights? Inverse vol, equal, cap, …

� Optimized: Minimum volatility, equal risk weight, …• Which covariance matrix?

• Sector neutral?

• Name constraints?

• Tracking error aware?

Similar issues with other Smart Betas

� Fundamental indexing• Which measure of value?

• Earnings, revenues, sales, cash flow, composite• Which time period?

� Momentum• Which return window?

• Exclude most recent period?• Scale returns by volatility?

• Which volatility?• Which weighting scheme?

� Quality• Which measure of quality?• Which weighting scheme

7

Evaluating the importance of Smart Beta

strategy design

� Simulate multiple versions of: Low Volatility,

Momentum and Fundamental strategies

� Evaluate how portfolio construction choices

affected simulated performance• Returns

− Full period, sub periods

• Risk

− Volatility, beta, tracking error

− Factor exposures

− Drawdowns: absolute, relative

• Sharpe Ratio

8

1. Low Volatility case study

� Metrics: standard deviation, residual risk,

beta• Calculated using 1, 3 & 5 year windows

� Universe: 500 largest US stocks

� All 500 stocks, 250 lowest risk, 100 lowest risk

� With and without sector neutrality constraint

� 25 years: 1989 – 2013, rebalanced quarterly

9

Low vol score depends on: Data window, total

vs residual risk, volatility vs beta

10Source: Dugar & Goldsticker, CRSP, Compustat, 500 largest stocks, data ending 12/31/2013

Time window, partitioning and sector neutrality

had little impact on overall low vol performance

Source: Dugar & Goldsticker, CRSP, Compustat, 500 largest stocks, data ending 12/31/2013 11

Low vol strategies underperformed pre, and

outperfomed post bursting of Tech Bubble


Differences in portfolio characteristics of low vol

strategies were not large

Cap

Weight

Change Window used to Calculate Vol Partition Sector

Netural,

3 Yr Vol1/(1 Yr Vol) 1/(3 Yr Vol) 1/(5 Yr Vol)250 Stocks

3 Yr Vol

100 Stocks

3 Yr Vol

Annual Return 11.34% 13.09% 12.92% 12.87% 11.97% 11.19% 12.96%

1989 - 1999 21.15% 17.27% 17.00% 16.85% 15.12% 14.18% 18.89%

2000 - 2013 4.73% 9.91% 9.82% 9.84% 9.55% 8.89% 8.51%

Volatility 16.1% 15.2% 15.4% 15.5% 13.3% 12.1% 16.0%

R-Square 94% 94% 94% 86% 76% 98%

Tracking Error 5.07% 5.31% 5.45% 6.84% 7.94% 3.45%

Sharpe Ratio 0.49 0.63 0.61 0.61 0.64 0.64 0.59

Beta 1.00 0.89 0.89 0.90 0.71 0.57 0.97

Up Beta 0.85 0.86 0.86 0.62 0.43 0.96

Down Beta 0.97 0.99 0.98 0.75 0.57 1.03

Max Drawdown -43% -42% -43% -43% -40% -35% -44%

Max Rel DD -38% -40% -41% -52% -56% -30%


Fama-French Factor Exposures

Market

Size

Small - Big

Value

High - Low Momentum

Cap Weighted 0.98 -0.16 0.01 -0.01

1 Year Volatility 0.89 0.00 0.27 -0.03

3 Year Volatility 0.90 -0.01 0.29 -0.05

5 Year Volatility 0.90 -0.01 0.30 -0.06

250 Stocks, 3 Year 0.76 -0.12 0.38 -0.02

100 Stocks, 3 Year 0.65 -0.21 0.37 0.01

Sector Neutral, 3 Yr 0.95 0.00 0.15 -0.02

Source: Dugar & Goldsticker, CRSP, Compustat, Ken French, 500 largest stocks, data ending 12/31/2013 14

Principle Component Exposures

PC 1:

Market

PC 2:

Momentum

PC 3:

???

PC 4:

Value

Cap Weighted 0.96 0.19 -0.04 0.09

1 Year Volatility 1.00 -0.03 0.03 0.01

3 Year Volatility 1.00 -0.04 0.03 0.03

5 Year Volatility 0.99 -0.05 0.03 0.04

250 Stocks, 3 Year 0.96 -0.28 0.01 -0.03

100 Stocks, 3 Year 0.87 -0.45 -0.03 -0.13

Sector Neutral, 3 Year 0.99 0.11 0.00 0.04


Low volatility case study findings:

Simulated returns:

� All of the low volatility strategies we examined exhibited similar patterns of returns.

Underperformed through 1999, outperformed since.

� Investing in only the least volatile stocks produced less volatile portfolios.

But, also lower returns, higher tracking error and larger drawdowns.

Comparison of metrics:

� Standard deviation, beta and tracking error produced similar results.

� 1, 3 and 5 year statistics produced similar results.

Correlation between 1 and 5 year vol is only .5, but dispersion is in higher vol stocks (low weights).

16

2. Fundamental Indexing Case Study

� Metrics: Earnings, Dividends, Sales, Book Value;

combination of the four

� Weighting scheme:

� (Stock Fundamental) / (Aggregate Fundamental)

� Look back: most recent 4, 8 & 12 quarters



� Note: Partitioning the universe and sector constraints were not investigated. They are inconsistent with the philosophy behind fundamental indexing.

17

Fundamental Indexing depends on metric and look-back

But shows less variability than other strategies


Similar performance of fundamental strategies

Book value diverged post Financial Crisis


Fundamental strategies underperformed pre,

and outperfomed post bursting of Tech Bubble


Characteristics of fundamentally weighted

strategies were similar

Cap Wt Book Earnings Revenues Dividends Comb 8Q

Annual Return 11.21% 10.70% 11.54% 11.71% 11.36% 11.44%

1989 - 1999 20.36% 17.77% 17.73% 17.40% 16.21% 17.67%

2000 - 2013 4.25% 5.23% 6.71% 7.26% 7.53% 6.59%

Volatility 16.2% 17.4% 15.5% 16.4% 15.0% 16.0%

R-Square 91% 92% 90% 81% 92%

Tracking Error 5.2% 4.3% 5.3% 6.6% 4.6%

Sharpe Ratio 0.48 0.42 0.52 0.51 0.53 0.50

Beta 1.00 1.02 0.91 0.95 0.83 0.94

Up Beta 1.02 0.92 0.97 0.82 0.95

Down Beta 1.04 0.89 1.01 0.76 0.93

Max Draw Down -44.3% -53.6% -47.9% -49.0% -50.1% -49.9%

Max Rel DD -21.7% -22.0% -24.5% -32.6% -22.4%



Market

Size

Small - Big

Value

High - Low Momentum

Cap Weighted 0.98 -0.16 0.01 -0.01

Book 1.01 -0.12 0.30 -0.16

Earnings 0.91 -0.18 0.23 -0.12

Revenue 0.94 -0.04 0.30 -0.11

Dividend 0.87 -0.25 0.37 -0.12

Comb 8Q 0.94 -0.15 0.28 -0.13



PC 1:

Market

PC 2:

Momentum

PC 3:

???

PC 4:

Value

Cap Weighted 0.96 0.19 -0.04 0.09

Book 0.97 0.04 0.02 0.22

Earnings 0.98 -0.03 0.00 0.19

Revenue 0.97 0.01 0.03 0.17

Dividend 0.95 -0.20 -0.01 0.17

Comb 8Q 0.98 -0.02 0.01 0.19


Fundamental Indexing case study findings:

24

Simulated returns:

� All simulations exhibited similar return patterns

• Underperformed through 1999

• Sharp recovery

• Generally, performed in line since

� Book Value had more severe drawdown in

Financial Crisis, and less recovery since.


� Generally independent of data window.

� While there are some differences, all metrics

tend to produce similar stock weights

3. Momentum case study

� Metrics: 5 month, 11 month & 23 month returns

• Periods end 1 month before portfolio construction

• Convert raw momentum values to Z-Score

� Weighting scheme:

• Z > 0: 1 + Z; Z < 0: 1/(1-Z); Z=0: 1


� All 500, 250 highest Z-Score, 100 highest Z-Score

� With and without sector neutrality constraint


25

Momentum weights depend on look-back

window and return vs Z-Score


More concentrated portfolios had superior long-

term performance, but also larger drawdowns


Outperformance of momentum primarily came

post bursting of Tech Bubble


Differences in portfolio characteristics of

momentum strategies were not large.

Cap

Weight

Window used to Calculate Momentum Partition 11 Month Sector

Netural,

11 Mth5 Month 11 Month 23 Month 250 Stocks 100 Stocks

Annual Return 11.34% 12.98% 14.17% 13.52% 12.83% 12.48% 13.8%

1989 - 1999 21.15% 17.72% 21.69% 20.97% 17.35% 17.01% 21.12%

2000 - 2013 4.73% 9.38% 8.59% 8.00% 9.39% 9.04% 8.30%

Volatility 16.1% 17.2% 17.5% 17.7% 17.4% 18.4% 17.2%

R-Square 95% 96% 97% 93% 88% 98%

Tracking Error 5.26% 4.62% 4.05% 6.40% 8.73% 3.69%

Sharpe Ratio 0.49 0.55 0.61 0.57 0.54 0.49 0.60

Beta 1.00 1.02 1.04 1.07 1.00 1.00 1.06

Up Beta 0.97 1.03 1.05 0.93 0.92 1.09

Down Beta 1.14 1.15 1.16 1.13 1.13 1.08

Max Drawdown -43% -44% -44% -45% -43% -46% -44%

Max Rel DD -37% -31% -30% -41% -48% -32%



Market

Size

Small - Big

Value

High - Low Momentum

Cap Weighted 0.98 -0.16 0.01 -0.01

5 Mth Momentum 1.00 0.09 0.21 0.02

11 Mth Momentum 1.04 0.13 0.09 0.13

23 Mth Momentum 1.05 0.06 0.06 0.09

250 Stocks, 11 Mth 1.04 0.18 0.04 0.28

100 Stocks, 11 Mth 1.08 0.30 -0.02 0.43

Sect Neutral, 11 Mth 1.02 0.07 0.06 0.09



PC 1:

Market

PC 2:

Momentum

PC 3:

???

PC 4:

Value

Cap Weighted 0.96 0.19 -0.04 0.09

5 Mth Momentum 0.98 0.15 0.01 -0.03

11 Mth Momentum 0.96 0.27 -0.02 -0.08

23 Mth Momentum 0.96 0.24 -0.01 -0.05

250 Stocks, 11 Mth 0.91 0.35 -0.07 -0.18

100 Stocks, 11 Mth 0.83 0.45 -0.12 -0.27

Sect Neutral, 11 Mth 0.96 0.26 -0.02 -0.04


Momentum case study findings

32

Simulated returns:� All of the momentum strategies we examined

exhibited similar patterns of returns

• Most of outperformance came post Tech Bubble

• Underperformance during Financial Crisis

• Outperformance since


� Full simulation: 11 month outperformed 5

� 5 month momentum worst performer before and during Tech Bubble.

� More concentrated portfolios had larger factor exposures and performance cycles.

Smart Beta factor exposures across sleeves

� Fama-French

� Principle Components

Source: 33


Market

Size

Small - Big

Value

High - Low Momentum

Capitalization

Weighted0.98 -0.16 0.01 -0.01

3 Year Volatility 0.90 -0.01 0.29 -0.05

100 Lowest 3 Year

Volatility0.65 -0.15 0.37 0.04

Fundamental

Indexing0.94 -0.15 0.28 -0.13

11 Month

Momentum1.04 0.13 0.09 0.13

100 Highest 11 Mth

Momentum1.08 0.30 -0.02 0.43


Principle Components Analysis

PC 1:

Market

PC 2:

Momentum

PC 3:

???

PC 4:

Value

Capitalization

Weighted0.96 0.19 -0.04 0.09

3 Year Volatility 1.00 -0.04 0.03 0.03

100 Lowest 3 Year

Volatility0.87 -0.45 -0.03 -0.13

Fundamental

Indexing0.98 -0.02 0.01 0.19

11 Month

Momentum0.96 0.27 -0.02 -0.08

100 Highest 11 Mth

Momentum0.83 0.45 -0.12 -0.27


Summary of simulated returns

� For each Smart Beta, results were largely

insensitive to the portfolio construction rules.• Average returns

• Drawdowns

• Behavior during Tech Bubble and Financial Crisis

� 100 and 250 stock portfolios• Larger drawdowns and cycles of relative

performance

• More extreme factor exposures

� Sector neutral portfolios• Less tracking error

Source: 36

Conclusions:

Pick the approach for economic reasons.� Don’t rely on backtests.� Why do you want the exposure?

• What risk or behavior is it intended to capture?• Pick approach most closely aligned with objective.

� Consider exposures to unintended risks.

How does the strategy fit into your aggregate portfolio?� Factor exposures� Is its behavior complimentary?

Pay attention to other portfolio characteristics.� Capacity, liquidity and concentration� Turnover and transactions costs� Management fees

37

Questions?

38

Economy & Finance

How important are the rules used to create smart beta portfolios