NDR

Ned Davis
Research
Group
NDR Solutions Quarterly
Investment Insight from NDR's Custom Solutions Team
ASSET ALLOCATION
MODEL ATTRIBUTION: A CASE STUDY
JUNE 2015
LISA MICHALSKI, CFA Associate Director, Custom Research Solutions

MATT BAUER, CFA Senior Research Analyst, Custom Research Solutions
NED DAVIS RESEARCH GROUP
NDR Solutions Quarterly | JUNE 25, 2015
TABLE OF CONTENTS
Executive Summary......................................................................................................................................................... 1
Introduction........................................................................................................................................................................ 2
Case Study: Global Balanced Account Model.................................................................................................. 2-3
Real-Time Performance: October 31, 2012 - May 31, 2015........................................................................... 4
Measuring Individual Asset Contribution............................................................................................................ 5
Setting Performance Expectations.......................................................................................................................... 6
Evaluating Real-Time Performance.......................................................................................................................... 7
Model and Sub-Model Decomposition........................................................................................................... 8-11
Indicator Analysis: Stock/Bond Externals.................................................................................................... 12-14
Testing Indicator Alternatives........................................................................................................................... 15-19
Conclusion......................................................................................................................................................................... 19
Appendix A: Back-Test Supplementary Analysis...................................................................................... 20-22
Appendix B: Real-Time Supplementary Analysis..................................................................................... 23-24
Appendix C: Stock/Bond External Model Alternatives.......................................................................... 25-26

Please see important disclosures at the end of this report.
www.ndr.com | Periodical | Issue #CRS201506251
This edition of NDR Solutions Quarterly discusses performance attribution for tactical asset allocation models. As
originally stated, the goal of this publication is to provide our client base with unique investment ideas each quarter that
stretch beyond Ned Davis Researchs traditional strategy content. Through this publication, the NDR Solutions team can
showcase the varied skill sets and diverse experience we regularly provide to clients with direct engagement, customized
output, and tailored solutions. As always, we welcome any feedback, as well as questions about our services, and hope
that you find this publication insightful, unique, and useful.
EXECUTIVE SUMMARY
Implementation of a performance attribution framework is essential for the proactive monitoring and maintenance of
quantitative models. The attribution process evaluates the performance of specific components in a models hierarchy at any
point in time. While much emphasis is traditionally placed on model development and historical returns, attribution determines how well a model meets expectations as we move farther away from its back-tested history. By identifying a models
successes as well as shortcomings, attribution maintains model transparency and targets areas for incremental improvement, potentially avoiding the expense of a complete reconstruction in the future.
Measuring individual asset contribution is the entry point for understanding the most important decisions
made by a model vs. its benchmark over time.
Back-tested statistics frame the baseline expectation for evaluating a models real-time performance.
We perform an attribution analysis on NDRs Global Balanced Account Model, a stock-bond-cash tactical asset
allocation model with 31 months of real-time history.
The Balanced Account Models real-time performance is, overall, in-line with the expectations set by the backtest. We identify and closely examine a four-month underperforming period in the summer of 2013.
We decompose the Global Balanced Account Model into its four sub-models and perform attribution on each.
Three of the four sub-models exhibited real-time performance in-line with back-test statistics.
Our objective analysis combined with subjective interpretation flags three indicators for potential revision. Of
three indicator alternatives tested, we select two for potential inclusion in the model at a later date.
Changes to the Global Balanced Account Model are not recommended at this time, but we will follow our
two indicator alternatives in tandem with their counterparts and perform a comparative analysis in six to 12
months.
While Eugene Fama is often thought of as the Godfather of factor returns decomposition in modern portfolio theory, it was a seemingly
innocuous article in the Financial Analysts Journal in the summer of 1986 by Brinson, Hood and Beebower that spurred popular interest
in attributing portfolio performance to specific elements. Nearly 30 years later, Determinants of Portfolio Performance continues to
inspire financial analysts to better understand the forces driving portfolio returns. In this vein, we apply our own attribution analysis to
an NDR flagship asset allocation model.
Quantitative timing models can be highly effective tools
for building a disciplined understanding of the asset allocation
landscape. Models make objective, data-driven decisions; they
are immune to the influence that dramatic market moves have
on investor behavior. A models performance history provides
insight into the risk and return characteristics of different
investment regimes. And when proactively monitored and
maintained, quantitative timing models can provide the asset
manager with years of valuable decision-making guidance.
This paper focuses on attribution analysis for real-time

performance monitoring of asset allocation models. We
emphasize real-time because the true value of a model comes
from the guidance it offers as we move forward, not backward,
in time. To state the obvious, no guarantees exist that a models
back-tested performance (deliberately designed to produce
meaningful excess return over a lengthy and diverse history)
will continue into the future. A properly executed attribution
framework helps the analyst navigate and appropriately
respond to the change that the future inevitably brings.
Performance attribution enables us to determine the

efficacy of the tactical decisions made by a model. Attribution
analysis can be applied to any time period in a models history
back-tested, real-time1 or both and at any component level:
top level model, sub-models, and even individual indicators.
By definition, attribution analysis promotes transparency and
ease of understanding at any level in a models hierarchy. This,
in turn, enables the asset manager to act with an appropriate
level of confidence when responding to the models readings.
In the pages that follow, we outline an asset allocation

attribution process that facilitates proactive model performance
monitoring, determines the effectiveness of components within
a models hierarchy, and identifies specific indicators in need
of increased attention. We will demonstrate our techniques by
means of a case study on a Ned Davis Research asset allocation
model. Lastly, we will test factor adjustments within a sub-model
flagged by the attribution analysis and set up a framework to
monitor these changes for potential use in the future.
Monthly Data 1990-01-31 to 2015-05-31 (Log Scale)
NDR Global Balanced Account Model

Equity Line
750
CASE STUDY: GLOBAL BALANCED

ACCOUNT MODEL
NDRGs Global Balanced Account
Model combines trend, sentiment,
valuation,
and
macroeconomic
indicators in a weight-of-the-evidence
approach to assign monthly tactical
asset weights vs. a fixed benchmark of
55% stocks, 35% bonds, and 10% cash.
The model was launched in the NDR
service in October 2012 (chart at right),
and no changes have been made to it
since then.
1. The real-time period for a model begins when
testing and development ceases and the model is
allowed to operate on data that has never before
been observed. Real-time is the only true outof-sample period for a model, regardless of any
efforts made to cross-validate indicators or ensure
robustness throughout the back-test history.
500
Model
Model
Benchmark
Benchmark
Gain/Annum
750
8.9%
6.5%
250
250
Back-Test
1990
1992
1994
1996
Benchmark Based On 55% Stocks, 35% Bonds, & 10% Cash

1998
2000
2002
2004
2006
100 Benchmark: MSCI All Country World TR Index
Model Equity %
2008
2010
2012
2015-05-31 = 55.7%
RealTime
2014
100
75
75
50
50
25
25
100 Benchmark: Barclays Global Aggregate TR Bond Index
Model Bond %
2015-05-31 = 34.8%
100
75
75
50
50
25
25
25
Benchmark: Equal-Weighted 3-month Treasury Composite TR
Model Cash %
2015-05-31 = 9.6%
25
20
20
15
15
10
10
NSQ2015Q1_01
500
All performance numbers do not include transaction costs.
Copyright 2015 Ned Davis Research, Inc. Further distribution prohibited without prior permission. All Rights Reserved.
See NDR Disclaimer at www.ndr.com/copyright.html For data vendor disclaimers refer to www.ndr.com/vendorinfo/
The construction of this model started with the evaluation of individual indicators to gauge market risk consistently throughout
time. Indicators passing a rigorous initial screening process were then combined into equal-weighted diffusion indices and re-evaluated
in terms of their aggregate explanatory power, with special attention paid to avoid high correlation between individual indicators.2
In all, 22 indicators were selected for final inclusion in the model, organized in a hierarchical structure as follows:
Global Balanced
Account Model
Stock/Bond
Relative Model
Stock/Bond
Internal Model
(5 Indicators)
Bond/Cash
Relative Model
Stock/Bond
External Model
(6 Indicators)
Bond/Cash
Internal Model
(5 Indicators)
Bond/Cash
External Model
(6 Indicators)
The overall model consists of two individual relative models: a stock-bond component, evaluating the relative attractiveness
of stocks vs. bonds, and a bond-cash component, evaluating the relative attractiveness of bonds vs. cash. Each of these
components, in turn, is an equal-weighted composition of 2 diffusion-based sub-models: one internal (tape-based) model
composed of five indicators and one external (valuation, sentiment, macroeconomic) model composed of six indicators for
each of the stock/bond and bond/cash pairs. A Black-Litterman optimizer determines the periodic weight recommendations
utilizing the end-of-month model readings along with the asset covariance matrix and risk tolerance parameters.
The Global Balanced Account Model is an ideal candidate for this case study for the following reasons:
Simplicity: The model consists of just three assets (stocks, bonds, and cash) and recommends tactical positions relative
to static benchmark weights. In addition, the model is long-only; the recommended weights must always sum to 100%; it
rebalances at the end of every month; and no out-of-benchmark assets are allowed.
Available History: No changes have been made to the model or its components since it launched on October 31, 2012.
Thus, it offers 31 months of real-time performance to compare with more than 22 years of back-tested history.
Organizational Structure: The hierarchy of the model, as outlined above, naturally lends itself to component isolation and
re-evaluation at any level.
Real-Time Excess Return: Thus far, the model has logged a positive track record in its out-of-sample real-time period (discussed in the next section). Therefore, we are able to invest the time to thoughtfully evaluate the components and test ideas
for future enhancements rather than introducing hasty fixes to address a situation of lagging returns.
2. Since this paper focuses on model performance analysis, the details of indicator development on both a stand-alone basis and in combination will not be covered. A
complete explanation of NDRs model building process is covered in Chapter 2 of Ned Davis book, Being Right or Making Money (Third Edition).
REAL-TIME PERFORMANCE: OCTOBER 31, 2012 MAY 31, 2015

The chart below isolates the Global Balanced Account
Models 31-month real-time performance period. Since October
31, 2012, the model generated an excess return over the
benchmark of 70 bps per annum. With the exception of a short
period in 2013, the model consistently over-weighted stocks
and under-weighted bonds and cash in real-time, as displayed
in the bottom three clips of the chart.
For comparison, we note that the back-tested historical excess
return from the models inception on January 31, 1990 through
October 31, 2012 was 250 bps/annum: more than three times the
real-time excess return. Its important to keep in mind, however,
that the 22-year back-test history included many opportunities
for the model to accumulate substantial gains versus the

benchmark: most notably the boom and bust of the stock market
bubble in the late 1990s, followed by the housing bubble of the
mid-2000s, and the credit crisis in the fall of 2008.
Conversely, the 31-month real-time market environment
has been relatively benign, and the model has not yet been
given the chance to prove its potential value in a stock market
downturn. Our initial assumption is 70 bps/annum excess return
in real-time compared to 250 bps/annum during the back-test
poses no immediate concern. Nevertheless, we want to better
understand the drivers behind these figures in preparation for
when market conditions do change.
NDR Global Balanced Account Model Real-Time Performance

130
125
Equity Line
Model
Model
Benchmark
Benchmark
Gain/Annum
130
11.6%
10.9%
125
120
120
115
115
110
110
105
105
100
100
Benchmark Based On 55% Stocks, 35% Bonds, & 10% Cash
Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May
2013
2014
2015
100
Benchmark: MSCI All Country World TR Index
Model Equity %
2015-05-31 = 55.7%
100
75
75
50
50
25
25
100
Benchmark: Barclays Global Aggregate TR Bond Index
Model Bond %
2015-05-31 = 34.8%
100
75
75
50
50
25
25
25
Benchmark: Equal-Weighted 3-month Treasury Composite TR
Model Cash %
2015-05-31 = 9.6%
25
20
20
15
15
10
10
NSQ2015Q1_02
4
MEASURING INDIVIDUAL ASSET CONTRIBUTION

In our asset allocation framework, the attribution process
begins with a simple calculation that measures the periodic
effectiveness of over- or underweighting an individual asset
relative to the other assets in the model. The following formula
applies:
=
!
!!!
Weight. If the asset is overweight, the result will be positive; if

underweight, negative; and the result will be zero if the model
weight exactly equals the benchmark weight. In February, the
model over-weighted stocks by 5.75%, underweighted bonds
by -1.1%, and under-weighted cash by -4.65% (4th column
below).
! ! !
Next, we compute the Asset Contribution by multiplying

the Overweight vs. Benchmark by the Asset Return observed
in that holding period. For the month of February, stocks gained
5.62%, bonds fell by -0.591%, and cash gained nothing. The
resulting contribution for each asset appears in the rightmost
column in the table.
where r is the periodic excess return of the model over its

benchmark, i represents an individual asset in the model
(comprised of n assets) and Returni is the periodic return for
asset i.3
In the following example, we compute the contribution of
each asset in the model for the month of February 2015. A stepby-step application of the formula appears in the table below.
Finally, we calculate the overall model excess return by

summing the Asset Contribution for all assets. In the month of
February, the model realized an excess return of 0.33% (0.323%
contribution from stocks, 0.007% from bonds, and 0% from cash).
First, we determine the Overweight vs. Benchmark for each

asset by subtracting the Benchmark Weight from the Model
MODEL EXCESS RETURN AND INDIVIDUAL ASSET CONTRIBUTION FOR FEBRUARY 2015
Asset
Model Weight (%)
Benchmark
Weight (%)
Overweight vs.
Benchmark (%)
Asset Return (%)
Asset
Contribution (%)
Stocks
60.75
55
5.75
5.62
0.323
Bonds
33.9
35
-1.1
-0.591
0.007
Cash
5.35
10
-4.65
Model Excess Return (sum):

Ned Davis Research Group
0.33%
T_NSQ201506251.1
3. This is the Market Allocation Contribution component of the attribution formula introduced in Study Session 17 of the CFA Level III curriculum.
SETTING PERFORMANCE EXPECTATIONS

We extended the single-period attribution formula to
the model's 22 years of back-tested history, generated a
time-series for each assets contribution, and plotted them
alongside the models periodic excess returns (chart below).
Black dashed horizontal lines indicate the number of standard
deviations from the mean from the model's inception date
through October 31, 2012.
We also plotted the distribution of the model's excess

returns (see Appendix A) and noted that of 273 observations
in the back-test, 89% were within +/- one standard deviation
of the historical mean, 5% fell below -1 standard deviation,
and 5.5% were above +1 standard deviation. The distribution
is positively skewed, with a mean excess return of 0.19% and
median excess return of 0.053%.
In the back-tested history, the overall model excess returns

(top clip) have a 97% correlation with the stock contribution (2nd
clip), a 24% correlation with the bond contribution (3rd clip), and
essentially no correlation with the cash contribution (4th clip).
This tells us that the model's future success relies on a correct
tactical allocation to stocks more so than any other asset.
This baseline analysis provides a frame of reference to

determine whether the models real-time performance aligns
with our expectations. In the next section, we will compare
these back-test statistics to our 31 months of real-time data.
Monthly Excess Return of GBA Model and Assets During Back-Test History
Average: 0.19%
Std. deviation: 0.68%
Risk-adjusted ratio: 0.29
Model Monthly Excess % Return
5
4
3
2
1
0
-1
-2
Monthly Data 1990-02-28 to 2012-10-31

5
4
3
2
1
0
-1
-2
+2 SD
+1 SD
Mean
-1 SD
-2 SD
5
4
3
2
1
0
-1
-2
Stock Monthly Excess % Return
correlation with model returns: 0.97
Average: 0.17%
Bond Monthly Excess % Return

(4x Scale)
Average: 0.03%
Cash Monthly Excess % Return

(32x Scale)
correlation with model returns: -0.04
Average: -0.01%
Risk-adjusted ratio: -0.47
5
4
3
2
1
0
-1
-2
+2 SD
+1 SD
Mean
-1 SD
-2 SD
1.25
1.00
0.75
0.50
0.25
0.00
-0.25
-0.50
+2 SD
+1 SD
Mean
-1 SD
-2 SD
0.15
0.10
1.25
1.00
0.75
0.50
0.25
0.00
-0.25
-0.50
0.15
0.10
0.05
0.00
-0.05
0.05
+2 SD
+1 SD
Mean
-1 SD
-2 SD
0.00
-0.05
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
NSQ2015Q1_03
See NDR Disclaimer at www.ndr.com/copyright.html For data vendor disclaimers
refer to www.ndr.com/vendorinfo/
|
|
www.ndr.com
Periodical
Issue #CRS201506251
EVALUATING REAL-TIME PERFORMANCE

As discussed earlier, the Global Balanced Account Model has outperformed
its benchmark thus far in the 31-month
real-time period from October 31, 2012
through May 31, 20154, producing 70 bps
per annum in excess return. While the
out-of-the-gate performance is encouraging, we are hesitant to take it at face
value. Instead, we evaluate the models
real-time behavior in the context set by
the back-test.
The chart at right presents the identical
asset contribution breakdown shown on
page six but for the real-time period
only. The mean and standard deviation
dashed horizontal lines are sourced from
the back-test and serve as the guardrails
to evaluate the real-time performance.
We
highlight
the
following
observations from the real-time period:
The stock contribution has a 98%
correlation with model excess returns, consistent with the 97% correlation from the back-test period.
Bond contribution correlation is
43%, higher than the back-test,
and cash contribution correlation is
-0.09%, roughly the same.
Roughly 74% of the models real-time

monthly excess returns reside within
+/- 1 standard deviation of the realtime mean (see Appendix B), lower
than the back-test measure of 89%.
The real-time distribution is negatively
skewed, unlike the positively skewed
back-test excess returns.
expectations. We recognize, however,

that the proportion of excess return
outlier events observed in real-time
outnumber those from the back-test
2-to-1. This, along with the real-time
distribution's negative skew and summer
2013 underperformance inspires a
deeper dig into the model's components.
Overall, the
performance is
4. The last trading day of May 2015 was the 29th.
The models real-time excess return

neared the -1 standard deviation
boundary in July 2013 and January
2014, driven by an incorrect allocation
to stocks (top two clips, highlighted).
Monthly Data 2012-11-30 to 2015-05-29
Monthly Excess Return of GBA Model and Assets in Real-Time

1.5
1.0
Model Monthly Excess % Return

Back-test average: 0.19%

Back-test std. deviation: 0.68%
Back-test risk-adjusted ratio: 0.29
+1 SD
Average: 0.06% 1.5

Risk-adjusted ratio: 0.31 1.0
0.5
0.0
-0.5
0.5
Mean
0.0
-0.5
-1 SD
Standard deviation lines based on the back-test.
1.5
Stock Monthly Excess % Return
1.0

Average: 0.07% 1.5

Risk-adjusted ratio: 0.37 1.0
+1 SD
0.5
0.0
-0.5
0.5
Mean
0.4
0.2
0.0
-0.5
-1 SD
Bond Monthly Excess % Return

(4x Scale)
0.3
The models mean excess return (top

clip) is much lower in real-time than in
the back-test (0.06% vs. 0.19%); however, the standard deviation of excess
returns during the real-time period
is also much lower than in the backtest (0.19% vs. 0.68%). Computing a
risk-adjusted ratio (mean/std. deviation) yields a similar result for both the
back-test and the real-time period.
real-time model
in-line with our

+1 SD
Average: 0.0% 0.4

Risk-adjusted ratio: -0.12 0.3
0.2
0.1
0.0
-0.1
0.1
Mean
0.0
-0.1
-1 SD
-0.2
0.05
Cash Monthly Excess % Return

(32x Scale)
0.04
correlation with model returns: -0.09
0.03
Back-test average: -0.01%

Back-test risk-adjusted ratio: -0.47
-0.2
Average: 0.0% 0.05
Std. deviation: 0.0% 0.04
Risk-adjusted ratio: -0.96
0.03
0.02
0.01
0.00
-0.01
-0.02
0.02
0.01
+1 SD
0.00
-0.01
Mean
-0.02
-1 SD
Dec
Feb Apr
2013
Jun
NSQ2015Q1_04
Aug
Oct
Dec
Feb Apr
2014
Jun
Aug
Oct
Dec
Feb Apr
2015
See NDR Disclaimerwww.ndr.com
at www.ndr.com/copyright.html
For data vendor disclaimers
refer to#CRS201506251
www.ndr.com/vendorinfo/
| Periodical
| Issue
MODEL AND SUB-MODEL DECOMPOSITION

As outlined on page 3, the Global Balanced Account Model
consists of two relative models: one for stocks vs. bonds and
one for bonds vs. cash. To measure the tactical effectiveness of
these components, we must evaluate each one in a stand-alone
context. This is exactly the reverse of the model construction
process: instead of starting at the individual indicator level
and working our way up, we begin with the top-level models
and work our way down the hierarchy.
top-level model performance during the real-time period from

October 31, 2012 through May 31, 2015.
The isolated stock/bond top-level model outperformed
the 50/50 benchmark by 140 bps per annum in real-time by
over-weighting stocks relative to bonds during the majority
of this upward-trending period. This result was expected,
since we have already demonstrated that the parent Global
Balanced Account Model could not have outperformed in
real-time without a successful tactical allocation to stocks.
We first create a stand-alone tactical asset allocation

framework for the stock/bond top-level model and each of its
two sub-models, ignoring cash, and compare the result to an
equal-weighted benchmark of stocks and bonds. We remove
the Black/Litterman optimization and directly map the model
scores, which range from 0 (all indicators favor bonds) to 100
(all indicators favor stocks) with 50 as neutral, to asset weights
in the model.5 The chart below shows the isolated stock/bond
The bottom clip of the chart shows the periodic model

excess return during the real-time period, with dashed lines
indicating the median and -1 standard deviation thresholds
from the back-test period. The stock/bond top-level model
experienced two negative outlier events in real-time (July
2013 and January 2014), and it failed to achieve the historical
median excess return a maximum of four months in a row.
Stock/Bond Top-Level Model Real-Time Performance

130
125
120
Model
Benchmark
130
125
120
115
115
110
110
105
105
100
Title
Gain/Annum
Model
11.9%
Benchmark 10.5%
95
Std Dev
4.7%
4.0%
Downside Dev
2.5%
2.5%
Bat Avg
71.0%
Sharpe
2.54
2.62
Info Ratio
1.05
Tracking Err
1.3%
100
Max Drawdown
-2.0% (2013-05-31..2013-06-29)
-2.1% (2013-04-30..2013-06-29)
95
2013
2014
2015
100
Model Equity %
2015-05-29 = 51.7%
100
75
75
50
50
25
25
1.25
1.00
0.75
0.50
0.25
0.00
-0.25
-0.50
-0.75
-1.00
Stock Contribution
Bond Contribution
Positive Model Excess Returns
Above median excess return
Negative outlier excess return
Negative Model Excess Returns
1.25
1.00
0.75
0.50
0.25
0.00
-0.25
-0.50
-0.75
-1.00
Monthly Excess Return of Stock/Bond Top-Level Model

NSQ2015Q1_05
See NDR
Disclaimer
www.ndr.com/copyright.html
For data
vendor
disclaimersbe
refer
to www.ndr.com/vendorinfo/
5. A 50/50 benchmark enables us to directly map the model scores to asset
weights
andatremove
any performance that
would
otherwise
attributed
to Black/Litterman
optimization. In this scenario, a neutral score of 50 perfectly maps to a 50/50 model allocation with excess return of zero. The higher the score above 50, the higher (lower)
the model's allocation to stocks (bonds); the lower the score below 50, the lower (higher) the model's allocation to stocks (bonds).
Continuing down the stock/bond hierarchy, we perform

the same exercise with each of the internal and external
sub-models. The chart below shows the isolated stock/bond
internal model performance during the real-time period from
October 31, 2012 through May 31, 2015.
internal model compared to 140 bps/annum excess return in

the top-level model.
The stock/bond internal model experienced just one
negative outlier event in real-time (January 2014), and it failed
to achieve its historical median excess return for at most four
months in a row.
The isolated stock/bond internal model outperformed

the 50/50 benchmark by 320 bps per annum in real-time by
consistently over-weighting stocks during the majority of this
upward-trending period. The internal models stock allocations
(2nd clip) were more aggressive than the top-level models
stock allocations, resulting in increased outperformance over
the 50/50 benchmark: 320 bps/annum of excess return in the
The internal model is performing in-line with our expectations.

In fact, a model comprised of only technical indicators, as this
one is, should recognize and stay on the correct side of a steady
upward trend like the one observed in the past few years. We
expect the internal model to present more interesting data when
market conditions eventually change.
Stock/Bond Internal Model Real-Time Performance

135
130
125
Model
Benchmark
135
130
125
120
120
115
115
110
110
105
105
100
Title
Gain/Annum
Model
13.7%
Benchmark 10.5%
95
Std Dev
5.5%
4.0%
Downside Dev
2.4%
2.5%
Bat Avg
71.0%
Sharpe
2.50
2.62
Info Ratio
1.49
Tracking Err
2.1%
100
Max Drawdown
-2.1% (2013-05-31..2013-06-29)
-2.1% (2013-04-30..2013-06-29)
95
2013
2014
2015
100
Model Equity %
2015-05-29 = 70.0%
100
75
75
50
50
25
25
3.0
Stock Contribution
Bond Contribution
2.5
3.0
2.5
2.0
2.0
1.5
1.5
1.0
1.0
0.5
0.5
0.0
0.0
-0.5
-0.5
-1.0
-1.0
-1.5
-1.5
Monthly Excess Return of Stock/Bond Internal Model

NSQ2015Q1_06
|
|
www.ndr.com
Periodical
Issue #CRS201506251
Rounding out the sub-model analysis, we apply the same

approach to the stock/bond external model. The following
chart shows the stock/bond external isolated model
performance during the real-time period from October 31,
2012 through May 31, 2015.
The isolated stock/bond external model underperformed
the 50/50 benchmark by -40 bps per annum in real-time with
an indecisive allocation to stocks throughout this upwardtrending period. The external models stock recommendations
(2nd clip) remained within 33% of the 50% benchmark

allocation, fluctuating between over- and underweight every
few months and resulting in flat-to-down real-time relative
performance.
The stock/bond external model experienced two negative
outlier events in real-time (July 2013 and February 2015), and
it failed to achieve its historical median excess return for 13
months in a row (March 2014 March 2015).
Stock/Bond External Model Real-Time Performance

125
120
Model
Benchmark
125
120
115
115
110
110
105
105
100
Title
Gain/Annum
Std Dev
Downside Dev
Bat Avg
Sharpe
Info Ratio
Tracking Err
10.1%
4.1%
2.6%
51.6%
2.44
-0.29
1.5%
Benchmark 10.5%
4.0%
2.5%
Model
95
2.62
100
Max Drawdown
-2.0% (2013-04-30..2013-06-29)
-2.1% (2013-04-30..2013-06-29)
95
2013
2014
2015
100
Model Equity %
2015-05-29 = 33.3%
100
75
75
50
50
25
25
Stock Contribution
Bond Contribution
1.0

1.0
0.5
0.5
0.0
0.0
-0.5
-0.5
-1.0
-1.0
-1.5
-1.5
Monthly Excess Return of Stock/Bond External Model

NSQ2015Q1_07
See NDR Disclaimer
data vendor disclaimers
| Periodical
| Issue #CRS201506251
www.ndr.com
10 at www.ndr.com/copyright.html For
The table below summarizes the real-time decomposition

analysis for the stock/bond model and each of its two sub-models.
Though we expect an external model comprised of

sentiment, valuation, and macroeconomic indicators to behave
differently than its price-based counterpart, our analysis
suggests that the stock/bond external model warrants further
investigation. In the next section, we will continue down the
hierarchy of the stock/bond external model, examining each
of its six indicators in varying detail.
The external model tempered the aggressiveness of the

internal model during the real-time period, as displayed in the
Excess Gain/Annum vs. Benchmark column. However, the
top-level model beat or tied each of its sub-components in
both risk-adjusted terms as well as time spent outperforming
the benchmark, as shown in the Sharpe Ratio and % Months
Positive Excess Return columns.
STOCK/BOND ISOLATED MODEL REAL-TIME STATISTICS
Sharpe
Ratio
% Months
Positive
Excess
Return
Number of
negative
outlier
events
Max number of consecutive

months missing median
excess return
140 bps
2.54
71
Stock/Bond Internal Model
320 bps
2.50
71
Stock/Bond External Model
-40 bps
2.44
51.6
13
Excess Gain/
Annum vs.
Benchmark
Stock/Bond Top-Level
(50% Internal 50% External)
Model
T_NSQ201506251.2
We also performed the identical model isolation exercise

for the bond/cash model and each of its two sub-models.
The results are presented in the table below. The bond/cash
model and each of its two sub-models have consistently

outperformed a 50/50 benchmark of bonds and cash
throughout the real-time period.
BOND/CASH ISOLATED MODEL REAL-TIME STATISTICS
Sharpe
Ratio
% Months
Positive
Excess
Return
Number of
negative
outlier
events
Max number of consecutive

months missing median
excess return
120 bps
1.4
64.5
Bond/Cash Internal Model
130 bps
1.24
64.5
Bond/Cash External Model
120 bps
1.58
67.7
Excess Gain/
Annum vs.
Benchmark
Bond/Cash Top-Level
(50% Internal 50% External)
Model

T_NSQ201506251.3
11
INDICATOR ANALYSIS: STOCK/BOND EXTERNALS

As shown in the prior section, the Stock/Bond external
isolated model lagged its 50/50 benchmark by -40 bps
per annum during the real-time period from October 31,
2012 through May 31, 2015. The external model's lagging
performance in the summer of 2013 provides the opportunity
to examine the indicators in detail to better understand why
the model missed the mark.
The bottom panel of the chart lists the individual indicator

readings comprising the external stock/bond model score at the
end of each month. Internal and top-level model scores are also
shown for reference. Each of the six external indicators casts an
equal-weighted vote in favor of stocks, bonds, or in some cases
neutral, resulting in an external model score ranging from 0 (all
indicators favor bonds) to 100 (all indicators favor stocks).
The chart below shows the stock/bond external isolated

model for calendar year 2013. Both the model and benchmark
pushed higher after sideways movement throughout the
summer months. During that time, the external model took a
significant underweight position in stocks, as shown in the 2nd
clip. The underweight allocation lasted four months, resulting
in the model lagging the benchmark for the remainder of the
year.
The external model first turned bearish on stocks in May

2013, moving to 41.7 from Aprils neutral score of 50. The
external model grew even more bearish on stocks at the end
of June, posting a month-end score of 16.7, and it remained
bearish until the end of September when it turned neutral
once again. Model scores below 50 consequently result in
underweight allocations to stocks (overweight bonds), while
scores above 50 translate to overweight allocations to stocks
(underweight bonds).
Stock/Bond External Isolated Model

Model
Benchmark
108
108
105
105
102
102
Model Bond %
2013-12-31 = 33.3%
75
75
50
50
25
25
2013-12-31 = 66.7%
Model Equity %
U.S. Market Sentiment
Bonds
Stocks
Stocks
Bonds
Stocks
Stocks
Stocks
Stocks
Global Market Senitment
Bonds
Bonds
Neutral
Bonds
Bonds
Bonds
Bonds
Bonds
Stocks
Bonds
NDR Global SHUT RS
Stocks
Bonds
Bonds
Bonds
Bonds
Bonds
Stocks
Stocks
Stocks
Equity Risk Premium
Bonds
Bonds
Bonds
Bonds
Bonds
Bonds
Bonds
Bonds
Bonds
Central Bank Policy
Stocks
Stocks
Stocks
Stocks
Stocks
Stocks
Stocks
Stocks
Stocks
Global PMI Manufacturing
Stocks
Stocks
Bonds
Bonds
Bonds
Bonds
Bonds
Stocks
Stocks
External Model
50.0
50.0
41.7
16.7
33.3
33.3
50.0
66.7
66.7
Internal Model
60.0
60.0
50.0
50.0
60.0
60.0
70.0
70.0
80.0
Total Composite
55.0
55.0
45.8
33.3
46.7
46.7
60.0
68.3
73.3
Mar
Apr
Jun
Jul
Aug
Sep
Oct
Nov
Jan
2013
Feb
NSQ2015Q1_08
May
Dec
12
Individual indicator readings during the summer of 2013

(chart page 12, bottom clip) were as follows:
real-time period, as we have already done with the parent

model and sub-models.
U.S. Stock Market Sentiment changed to favor bonds

over stocks at the end of June and held the position for
one month, when it resumed favoring stocks over bonds.
The individual indicator isolated models produce more

dramatic asset weights than the parent models presented
earlier. Since individual indicators take at most one of three
positions (favoring stocks, bonds, or neutral), the isolated
indicator models take a 100% position in stocks when the
indicator favors stocks and a 100% position in bonds when
it favors bonds. Neutral readings translate to a benchmark
allocation of 50% stocks and 50% bonds.
Global Stock Market Sentiment favored bonds throughout the year, with the exception of a neutral reading posted at the end of May.
Relative Strength of the global Staples, Health Care,
Utilities, and Telecom (SHUT) sectors vs. the broad market changed to favor bonds over stocks at the end of April
and held this position throughout the summer until the
end of September, when it returned to favor stocks over
bonds.
The chart below shows the U.S. Market Sentiment standalone model during the real-time period from October 31,
2012 through May 31, 2015. The middle clip displays the U.S.
market sentiment time series along with standard deviation
brackets that generate the signals6 and is shaded blue when
the indicator favors stocks, black for bonds, and grey for
neutral.
The Equity Risk Premium indicator favored bonds the entire time.
Global Central Bank Policy favored stocks the entire time.
The isolated U.S. market sentiment model underperformed

14 allocation changes during this upward-trending period.
This model experienced three negative outlier events in realtime (July 2013, January 2014, and February 2015), and it
failed to achieve its historical median excess return for at most
five months in a row.
The Global Manufacturing Purchasing Managers Index

(PMI) changed to favor bonds over stocks at the end of
May and held this position through October, when it returned to favor stocks over bonds.
We will put this information to further use after we evaluate
each of the indicators on a stand-alone basis over the entire
Stock/Bond Model for U.S. Market Sentiment Real-Time Performance

125
120
Model
Benchmark
125
120
115
115
110
110
105
105
100
Title
Gain/Annum
Model
10.1%
Benchmark 10.5%
95
Std Dev
6.1%
4.0%
Downside Dev
3.8%
2.5%
Bat Avg
45.2%
Sharpe
1.65
2.62
Info Ratio
-0.10
Tracking Err
3.9%
Max Drawdown
-4.2% (2013-05-31..2013-08-31)
-2.1% (2013-04-30..2013-06-29)
100
95
2013
2014
2015
75
75
U.S. Stock Market Sentiment Composite 6-Month Smoothing

+0.25 Rolling 5-Week Standard Deviation
-2.0 Rolling 5-Week Standard Deviation
70
65
70
65
60
60
55
55
50
45
3
50
Model is 100% stock when 6-month smoothing crosses below lower bracket and reverses.
Stock Contribution
Bond Contribution
Model is 100% bond when 6-month smoothing crosses above upper bracket and reverses.

45
3
2
1
-1
-1
-2
-2
-3
-3
Monthly Excess Return of Stock/Bond Model for U.S. Market Sentiment

NSQ2015Q1_09
6. Discussion of signal generation techniques for specific indicators will only be covered on an as-needed basis so as not to drift too far off topic.
13
We performed the same analysis for the remaining five

external indicators on a stand-alone basis. The results are
presented in the table below.
Central bank policy remains accommodative, which is

favorable for stocks, and the indicators real-time call was
correct. The result of global easing, however, has worked to the
detriment of the equity risk premium trend indicator, which
remains out of favor partly due to unprecedented low bond
yields, driving investors to seek yield in riskier assets, pushing
equity prices (yields) higher (lower). We see no reason to tinker
with either of these indicators until we observe a return to
normal interest rate policy or conclude that a structural shift in
interest rate movements has taken place.
Not surprisingly, there is a wide dispersion of real-time

excess return among the individual indicators in the external
model. By design, the model employs a weight-of-the evidence
approach, allowing for individual indicators to be wrong for
periods of time and instead relying on the overall composite
score to point the tactical model allocations in a profitable
direction. For this reason, determining if an indicator is no
longer relevant or just temporarily out of favor requires
a subjective judgment call in addition to the objective
analysis we have applied to this point.
The Global Purchasing Managers Index (PMI) indicator

is behaving in accordance with our expectations, so we have
no immediate plans to make adjustments here either. The PMI
data only updates once per month and has a reporting lag, so
we expect this indicator to be slower-moving and support the
model from a macroeconomic perspective. Furthermore, the
indicator has outperformed in real-time, as shown in the table
(6th row).
To begin, we focus on the equity risk premium (4th row) and

global central bank policy (5th row) statistics presented in the
table. Upon first glance, it appears that the equity risk premium
indicator has been a total failure while the central bank policy
has been a complete success. While that is true in an objective,
stand-alone context, the model treats the indicators as a group,
so we need to consider the correlations they have with one
another before passing judgment on a single indicator.
Aside from completely redesigning or replacing these three

indicators, we dont believe that changes to them at this time
would materially improve the model performance in the future.
The equity risk premium and the central bank policy

indicators are thematically linked since they share an exposure
to a common factor: interest rates. Neither indicator has
changed its signal during the entire real-time period, so they
have been in perfect opposition with one another, canceling
out each others readings in the context of the parent model.
The three remaining external indicators, however, could

benefit from a little extra attention. In the remaining pages,
we test the impact of simple revision ideas for the U.S. market
sentiment, global market sentiment, and SHUT relative strength
indicators for potential substitution in the model at a later date.
EXTERNAL STOCK/BOND INDICATOR ISOLATED MODEL REAL-TIME STATISTICS

Number of
negative
outlier
events
Max number
of consecutive
months
missing median
excess return
Number
of signal
changes
Excess Gain/
Annum vs.
Benchmark
Sharpe Ratio
% Months
Positive Excess
Return
U.S. Stock Market Sentiment
-40 bps
1.66
46.7
14
Global Stock Market Sentiment
-670 bps
1.28
20
SHUT Index Relative Strength
160 bps
2.01
63.3
Equity Risk Premium
-710 bps
1.36
30
Global Central Bank Policy
730 bps
2.38
70
Global PMI
260 bps
2.11
60
Model

T_NSQ201506251.4
14
TESTING INDICATOR ALTERNATIVES

There is a fine line between chasing the past to build a better
back-test versus using what weve learned about a model in
real-time to test the impact of specific, incremental adjustments
for potential use in the future. In the latter scenario, we identify
what we perceive as reasonable changes to individual
indicators and then track these proposed changes outside the
context of the model in parallel over a suitable real-time period.
4. A more relevant or higher frequency alternative data set is

released.
5. Development of more robust signal generation techniques.
As analysts, we are continually evaluating the efficacy of our
methods and updating processes accordingly.
We conclude by testing revision ideas for the remaining
three stock/bond external indicators.
We would replace an indicator in the model only after a

convincing real-time analysis of a viable alternative, and in
doing so we would never overwrite performance history. Model
adjustments only apply toward the future, preserving the realtime statistics. To reiterate, our philosophy is geared toward
maintenance of the real-time model, not building a better
back-test.
As presented in the chart on page 13, the U.S. market

sentiment indicator logged adequate stand-alone performance
during the real-time period, underperforming the benchmark by
-40 bps/annum. With 14 allocation changes, it was the noisiest
indicator in the group, as shown in the table on page 14.
As we tracked the indicator in real-time, it appeared that the
bearish signals (favoring bonds over stocks) occurred too late at
times (December 2012, July 2013, March 2014, and January 2015)
and the oscillation between bearish and neutral readings was
too frequent. To address this potential concern, we adjusted the
signal-generating threshold for bearish signals only.
That being said, we do believe there are valid reasons to

change an indicator, including:
1. A permanent shift in the data that causes the indicator to
lose the effectiveness it once had.
2. A change in assumptions of what the indicator represents.
By raising the top bracket in the middle clip (chart below),

we shorten the timeframe required to generate a bearish signal
after sentiment reverses from a local high. The performance
during the real-time period appears in the chart below.
3. A desire to match the indicator more closely to the market its

measuring, such as cap-weighting versus equal-weighting.
Stock/Bond Model for U.S. Market Sentiment (New) Real-Time Performance

125
120
Model
Benchmark
125
120
115
115
110
110
105
105
100
Title
Model
95
Gain/Annum
8.5%
Benchmark 10.5%
Std Dev
Downside Dev
5.2%
2.7%
4.0%
2.5%
Bat Avg
35.5%
Sharpe
1.62
Info Ratio
-0.58
2.62
Tracking Err
3.5%
Max Drawdown
-4.2% (2013-05-31..2013-08-31)
-2.1% (2013-04-30..2013-06-29)
100
95
2013
2014
2015
U.S. Stock Market Sentiment Composite 6-Month Smoothing
+1.25 Rolling 5-Week Standard Deviation
-2.0Rolling 5-Week Standard Deviation
70
65
70
65
60
60
55
55
50
45
3
50
Model is 100% stock when 6-month smoothing crosses below lower bracket and reverses.
Stock Contribution
Bond Contribution
Model is 100% bond when 6-month smoothing crosses above upper bracket and reverses.
45
3
2
1
-1
-1
-2
-2
-3
-3
Monthly Excess Return of Stock/Bond Model for U.S. Market Sentiment (New)
NSQ2015Q1_10
15
Although the revised indicator generated improved backtested results over its history from 1981 (not shown), the
performance from October 31, 2012 through May 31, 2015
actually degraded to -200 bps/annum of excess return with 17
allocation changes, contrary to what we anticipated. Given the
indicators adequate performance in its original form, we dont
see an immediate need to test more radical adjustments at this
time. We will continue to watch the indicator going forward and
revisit it on an as-needed basis.
thresholds that generate the signals and is shaded blue when

the indicator favors stocks, black for bonds, and grey for neutral.
The isolated global market sentiment model underperformed
nine allocation changes during this upward-trending period.
This model experienced six negative outlier events in real-time,
and it failed to achieve the historical median excess return for a
maximum of eight months in a row.
The global sentiment data has a much shorter history than
its U.S. counterpart. With an inception date of December 31,
2002, the indicator had just 10 years of history when the model
launched in October of 2012. In real-time, this indicator took
nearly two years to generate a bullish signal, longer than any
other analogous period from the back-test. To address this,
we adjusted the brackets to better respond to the changing
volatility of the underlying data. The result appears in the
chart on page 17.
Unlike U.S. sentiment, the global sentiment indicator

appears in need of a more serious overhaul. The chart below
shows the global market sentiment indicator stand-alone
performance during the real-time period from October 31,
2012 through May 31, 2015.
The middle clip displays the deviation from trend line of the
global market sentiment index along with the static bracket
Stock/Bond Model for Global Market Sentiment Real-Time Performance

125
120
Model
Benchmark
125
120
115
115
110
110
105
105
100
Title
Gain/Annum
Model
3.8%
Benchmark 10.5%
95
Std Dev
3.1%
4.0%
Downside Dev
2.1%
2.5%
Bat Avg
19.4%
Sharpe
1.20
2.62
Info Ratio
-1.95
Tracking Err
3.4%
Max Drawdown
-3.5% (2013-04-30..2013-06-29)
-2.1% (2013-04-30..2013-06-29)
100
95
2013
2014
2015
Global Stock Market Sentiment Composite Deviation from Trend Line(5-Week/32-Week Smoothing)
1.4
1.4
1.2
1.2
1.0
1.0
0.8
0.8
0.6
0.4
3
0.6
Model is 100% stocks when deviation from trend line crosses below 0.5 and reverses.
Stock Contribution
Bond Contribution
Model is 100% bonds when deviation from trend line crosses above 1.25 and reverses.
0.4
3
2
-1
-1
-2
-2
-3
-3
Monthly Excess Return of Stock/Bond Model for Global Market Sentiment

NSQ2015Q1_11
16
In this case, our intuition worked to our benefit. In allowing

the brackets to adapt to the volatility of the series (2nd clip below),
the indicator has more opportunities to cast its vote, where in the
original version it was stuck in a bearish state for far too long. The
revised global market sentiment indicator still underperformed
the 50/50 benchmark in the real-time period but by a much
narrower margin than the original.
Lastly, we take a closer look at the global Staples, Health

Care, Utilities, and Telecom (SHUT) relative strength indicator.
Like U.S. sentiment, the SHUT indicator also posted decent
stand-alone performance during the real-time period. The
chart on page 18 shows the SHUT relative strength indicator
stand-alone performance from October 31, 2012 through May
31, 2015.
Stock/Bond Model for Global Market Sentiment (New) Real-Time Performance

125
120
Model
Benchmark
125
120
115
115
110
110
105
105
100
100
Title
95
Gain/Annum
Model
9.6%
Benchmark 10.5%
Std Dev
Downside Dev
Bat Avg
Sharpe
Info Ratio
Tracking Err
6.6%
4.0%
3.2%
2.5%
45.2%
1.44
2.62
-0.21
4.3%
Max Drawdown
-4.2% (2013-04-30..2013-08-31)
-2.1% (2013-04-30..2013-06-29)
95
2013
2014
2015
2.5
Global Stock Market Sentiment Composite Deviation from Trend Line(5-Week/32-Week Smoothing)
+0.75 Rolling 1-Year Standard Deviation 2015-06-19 = 1.166
-1.75 Rolling 1-Year Standard Deviation 2015-06-19 = 0.557
2.0
2.5
2.0
1.5
1.5
1.0
1.0
0.5
0.5
Model is 100% stocks when deviation from trend line crosses below -1.75 and reverses.
Stock Contribution
Bond Contribution
Model is 100% bonds when deviation from trend line crosses above 0.75 and reverses.
4
3
-1
-1
-2
-2
-3
-3
Monthly Excess Return of Stock/Bond Model for Global Market Sentiment (New)
NSQ2015Q1_12
17
The SHUT indicator generates a bearish signal (favoring

bonds over stocks) when the short-term moving average
trends above the long-term moving average, and a bullish
signal (favoring stocks over bonds) when the short-term
moving average trends below the long-term moving average
(2nd clip). Overall, weve been satisfied with the performance
of this indicator in real-time, but there were occasions when
the indicator took too long to reach the required threshold

to trigger a signal even though it was already on the correct
side of the trend. We reduced the margin required to generate
a new signal and introduced a neutral zone in the area in
between, as the indicator only posts bullish or bearish readings
in its current form. The result of these changes appears in the
chart on page 19.
Stock/Bond Model for Global SHUT Index Real-Time Performance

130
125
Model
Benchmark
130
125
120
120
115
115
110
110
105
105
The SHUT Index is composed of the Consumer Staples, Health Care, Utilities, and Telecom Sectors.
This largely defensive index helps distinguish market leadership as being either cyclical or defensive.
100
Title
Gain/Annum
Std Dev
Downside Dev
Bat Avg
Sharpe
Info Ratio
Tracking Err
12.1%
5.9%
3.6%
64.5%
2.04
0.37
4.4%
Benchmark 10.5%
4.0%
2.5%
Model
95
2.62
Max Drawdown
-3.4% (2013-12-31..2014-02-01)
-2.1% (2013-04-30..2013-06-29)
100
95
2013
2014
2015
140
140
SHUT Index/Global Broad Market Index Relative Strength 15-Day Smoothing

135
135
130
130
125
125
Model is 100% stock when 15-day smoothing crosses below 120-day smoothing by 1.5%
4
3
Stock Contribution
Bond Contribution
Model is 100% bond when 15-day smoothing crosses above 120-day smoothing by 1.5%.
4
3
2
-1
-1
-2
-2
-3
-3
Monthly Excess Return of Stock/Bond Model for Global SHUT Index

NSQ2015Q1_13
18
Our intuition also proved beneficial in this case, with

the adjusted SHUT relative strength indicator handily
outperforming its original version in the real-time period. As
we track this indicator going forward, we keep in mind the
potential for increased signal volatility as a result of narrowing
the bullish and bearish thresholds (2nd clip) and introducing a
neutral zone.
We prefer to work with small, incremental adjustments since

significant changes to the signal frequency of an individual
indicator can have a dramatic impact on correlations at the submodel level. Our goal is not to alter the model dramatically, but
instead to introduce reasonable improvements, where possible,
to keep it on track.
We also evaluated the revised global sentiment and SHUT
indicators at the stock/bond external sub-model level. These
three alternative external models (replacing global sentiment
only, SHUT only, and both) are presented in Appendix C.
The goal of this section has been to demonstrate the testing of

simple and logical ideas for alternative indicator choices, rather
than providing an exhaustive list of additional possibilities.
Stock/Bond Model for SHUT Market Index (New) Real-Time Performance

135
130
125
135
Model
Benchmark
130
125
120
120
115
115
110
110
105
100
Title
Gain/Annum
Std Dev
Downside Dev
Bat Avg
Sharpe
Info Ratio
Tracking Err
Max Drawdown
Model
12.7%
Benchmark 10.5%
95
5.3%
4.0%
1.6%
2.5%
58.1%
2.37
2.62
0.59
3.8%
-3.5% (2013-04-30..2013-08-31)
-2.1% (2013-04-30..2013-06-29)
105
100
95
2013
2014
2015
140
140

135
135
130
130
125
125
Model is 100% stock when 15-day smoothing crosses below 120-day smoothing by -0.25%
4
3
Stock Contribution
Bond Contribution
Model is 100% bond when 15-day smoothing crosses above 120-day smoothing by 0.25%
4
3
2
-1
-1
-2
-2
-3
-3
Monthly Excess Return of Stock/Bond Model for SHUT Market Index (New)
NSQ2015Q1_14
CONCLUSION
In the preceding pages, we have applied an objective, realtime attribution framework to an actual Ned Davis Research
asset allocation model, identifying the successful components
as well as those in need of increased attention. We performed
a full decomposition of the model and evaluated three ideas
for improvement at the individual indicator level.
Lastly, implementing small, incremental adjustments to a

model preserves the integrity of its real-time period, which is
the only true out-of-sample performance. The emphasis on
monitoring and maintenance also helps to avoid a complete
rebuild of the model in the future, since we focus on
manageable improvements one step at a time. Rebuilding
a model is a costly process not only in terms of the time
involved, but also in terms of convincing users of the prior
version to abandon it and adopt a brand new one having
no real-time history. The attribution process promotes
discipline and keeps us focused on these goals.
It is premature to recommend any changes to the model

at this time. Instead, we have created three shadow models to
monitor the real-time performance of our potential replacement
indicators. In doing so, new indicators have an opportunity to
earn a future position in the model. We view this as a systematic
alternative to introducing changes out of emotion or haste.
19
APPENDIX A: BACK-TEST SUPPLEMENTARY ANALYSIS

The distribution of excess returns from the back-test is
positively skewed (0.195% mean vs. 0.053% median), due
primarily to the positive outlier events in 2001, 2003, and 2008.
With such a large disparity between the mean and median
excess return, and given that we havent yet observed similar
market environments in real-time, its not reasonable to assume
that the model will achieve a mean monthly excess return of
0.195% over a relatively short real-time period. The historical
median return provides a suitable middle-of-the-road target to
measure the models consistency over time.
We identified consecutive periods of underperformance in

two ways:
1. monthly excess return less than the historical median
2. monthly excess return less than zero
In the first scenario it is still possible for the model to
outperform the benchmark, provided that the cumulative
excess return is greater than zero. In the second scenario, the
model underperforms the benchmark, due to the negative
excess return.
Data Range: -2.6 - 4.8
Distribution of Model Excess Returns During Back-Test Period
Stat
125
125
0.195
median
0.053
115
stDev
0.683
115
110
min
-2.561
4.757
110
105
max
# obs
273
105
skewness
2.884
kurtosis
19.027
100
120
100
95
Frequency
Value
mean
120
95
90
90
85
85
80
80
75
75
70
70
65
65
Number of Observations: 273

Number of Observations greater than 1 S.D.: 15 (5.49%)
Number of Observations less than 1 S.D.: 14 (5.13%)
Number of Observations within 1 S.D.: 244 (89.38%)
60
55
50
60
55
50
45
45
40
40
35
35
30
30
25
25
20
20
15
15
10
10
-2.5
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
Excess Return(%)
NSQ2015Q1_15
| Issue #CRS201506251
www.ndr.com | Periodical
20
The table at right displays the

number of consecutive months
during the back-test history that
the model missed expectations,
as defined on page 20.
In the back-test history, the
longest period the model failed
to record a positive excess
return lasted four months. This
happened once during the
22-year back-test history, as
displayed in the third column at
right (highlighted). The model
failed to achieve positive excess
return for three months in a
row in eight instances over the
course of the back-test history
(3rd column, 2nd row). Based on
these observations, a cause for
concern in real-time would be
an event when the model misses
positive excess return for five or
more months in a row.
Next, we turn to the median
return (2nd column of table), which
yields more interesting results.
In the back-test history, the
longest period when the model
consecutively missed the median
return lasted 8 months. This
happened twice (highlighted).
The model performance during
the most recent instance, October
2009 May 2010, is shown in the
chart at right.
During this eight-month
stretch, the model logged one
month of positive excess return
and seven months of negative
excess return, the last of which
was a negative outlier event in
excess of -1 standard deviation.
The following month, however,
the model recovered and hit the
median excess return (not shown).
Overall underperformance during
this period: 80 bps.
Number of
Consecutive Months
Model missed median

excess return
Model missed positive

excess return
2
3
4
5
6
7
8
9
10
11
12
12
9
6
1
1
1
2
0
0
0
0
16
8
1
0
0
0
0
0
0
0
0
T_NSQ201506251.5
Not Meeting Median Excess Return

107
107
Model
Benchmark
106
106
105
105
104
104
103
103
102
102
101
101
100
100
Title
99
Return %
Std Dev
Downside Dev
Bat Avg
Sharpe Info Ratio
Tracking Err
2.4%
2.5%
1.9%
14.3%
0.96
-3.8%
Benchmark 3.2%
2.5%
1.5%
Model
Oct
0.3
0.2
0.1
0.0
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
Nov
Stock Contribution
Dec
Cash Contribution
Max Drawdown
99
-4.4% (2010-04-30..2010-05-29)
1.27
Jan
2010
Bond Contribution
0.21
-3.8% (2010-04-30..2010-05-29)
Feb
Mar
Apr
May
0.3
0.2
0.1
0.0
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
Monthly Excess Return of GBA Model and Assets

NSQ2015Q1_16
21
The model also logged one sevenmonth and one six-month instance
during the back-test history when it
failed to achieve the median excess
return. These time periods, October 2001
April 2002 and March 2004 August
2004, are presented in more detail in the
charts at right and below.
Model
Benchmark
104
In the 2001-02 case (right), the

model logged two months of positive
(small) excess return and five months of
negative excess return with one negative
outlier event. Overall underperformance
during this period: -120 bps.
103
103
102
102
101
101
100
100
Title
In the 2004 case (below), the model

logged one month of positive excess
return and five months of negative
excess return with no negative outlier
events.
Overall
underperformance
during this period: -30 bps.
Return %
Model
1.5%
Benchmark 2.7%
99
Oct
Std Dev
Downside Dev
Bat Avg
Sharpe
Info Ratio
Tracking Err
1.1%
1.5%
0.9%
0.9%
33.3%
0.55
1.26
0.36
-3.5%
Nov
Stock Contribution
0.50
Dec
Bond Contribution
0.25
Our data confirms that the models

largest trouble spots during the backtest occurred when consistent periods
of underperformance also included
negative outlier events.
104
Jan
2002
Cash Contribution
Max Drawdown
-2.1% (2002-03-28..2002-05-01)
-2.0% (2002-03-28..2002-05-01)
Feb
99
Mar
Apr
0.50
0.25
0.00
0.00
-0.25
-0.25
-0.50
-0.50
-0.75
-0.75
-1.00
-1.00

NSQ2015Q1_17
See NDR Disclaimer atMonthly
www.ndr.com/copyright.html
For data vendorto
disclaimers
Data 2004-03-31
2004-08-31
(Log Scale)

Model
Benchmark
100
100
99
99

Title
Return %
Std Dev
Downside Dev
Model
-0.6%
0.6%
0.7%
Benchmark
-0.3%
0.5%
0.6%
Apr
0.3
0.2
Stock Contribution
Bat Avg
20.0%
Sharpe
-1.80
May
Bond Contribution
Tracking Err
-2.7%
Cash Contribution
Max Drawdown
-1.5% (2004-06-30..2004-07-31)
-1.3% (2004-06-30..2004-07-31)
Jun
0.1
Info Ratio
0.10
-1.47
Jul
Aug
0.3
0.2
0.1
0.0
0.0
-0.1
-0.1
-0.2
-0.2
-0.3
-0.3
-0.4
-0.4
-0.5
-0.5
-0.6
-0.6

NSQ2015Q1_18
22
APPENDIX B: REAL-TIME SUPPLEMENTARY ANALYSIS

Negative outlier events in excess of -1 real-time standard deviation occurred in 4 out of 31 cases, roughly 13% of the time:
more than double that of the back-test. Positive outlier events
were equally common, occuring in 4 out of 31 observations.
While the distribution of back-tested excess returns is positively skewed, the distribution of real-time excess returns is
negatively skewed (0.06% mean return vs. 0.074% median
return), due to the magnitude of the negative outlier events.
Data Range: -0.4 - 0.5

8.50
Distribution of Model Excess Returns During Real-Time Period

8.50
8.25
Stat
Number of Observations: 31
Number of Observations greater than 1 S.D.: 4 (12.90%)
Number of Observations less than 1 S.D.: 4 (12.90%)
Number of Observations within 1 S.D.: 23 (74.19%)
8.00
7.75
7.50
7.25
7.00
6.75
Frequency
6.50
Value
8.25
mean
0.06
median
0.074
stDev
0.19
7.75
min
-0.409
7.50
max
0.452
7.25
# obs
31
skewness
-0.673
7.00
kurtosis
3.826
8.00
6.75
6.50
6.25
6.25
6.00
6.00
5.75
5.75
5.50
5.50
5.25
5.25
5.00
5.00
4.75
4.75
4.50
4.50
4.25
4.25
4.00
4.00
3.75
3.75
3.50
3.50
3.25
3.25
3.00
3.00
2.75
2.75
2.50
2.50
2.25
2.25
2.00
2.00
1.75
1.75
1.50
1.50
1.25
1.25
1.00
1.00
-0.40 -0.35 -0.30 -0.25 -0.20 -0.15 -0.10 -0.05
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
Excess Return(%)
NSQ2015Q1_19
23
In terms of performance consistency

in real-time, the model has not yet
experienced any periods with two
consecutive months of negative excess
return (third column, table at right).
Number of
Consecutive Months
Model missed historical

median excess return
Model missed positive

excess return
In the context of historical median

return, there was one period when the
model missed its target for a five-month
stretch (2nd column, 4th row). The model
performance during this period, May
2013-September 2013, is shown in the
chart below.
During this five-month period, the

model logged two months of positive
excess return and three months of
negative excess return, with the July
2013 excess return nearing -1 historical
standard deviation (as set by the backtest). Overall underperformance during
this period: -40 bps.
T_NSQ201506251.6

Model
Benchmark
101
101
100
100
99
99
98
98
Title
Return %
Model
1.3%
Benchmark 1.7%
0.2
Downside Dev
0.6%
0.6%
Bat Avg
50.0%
Jun
May
2013
0.3
Std Dev
1.2%
1.4%
Stock Contribution
Bond Contribution
Sharpe
1.06
1.24
Info Ratio
0.21
Tracking Err
-1.8%
Max Drawdown
-1.9% (2013-05-31..2013-06-29)
-2.0% (2013-05-31..2013-06-29)
Jul
Cash Contribution
Aug

0.1
Sep
0.3
0.2
0.1
0.0
0.0
-0.1
-0.1
-0.2
-0.2
-0.3
-0.3
-0.4
-0.4
-0.5
-0.5
-0.6
-0.6

NSQ2015Q1_20
24
APPENDIX C: STOCK/BOND EXTERNAL MODEL ALTERNATIVES

Substituting the new global sentiment indicator into the
stock/bond external model (chart below) would have increased
the performance in the model's real-time period by 100 bps/year
over the original version; however, this comes with increased
standard deviation and a downside deviation increase of 60
bps/year. Similar to the original stock/bond external model,

this version also experienced two negative outlier excess return
events: July 2013 (identical to original model) and January 2014
(different from the original model).
Stock/Bond External Model With New Global Market Sentiment Indicator Real-Time Performance
130
125
130
Model
Benchmark
125
120
120
115
115
110
110
105
105
100
Title
Gain/Annum
Model
11.1%
Benchmark 10.5%
95
Std Dev
Downside Dev
Bat Avg
Sharpe
Info Ratio
Tracking Err
4.7%
4.0%
3.2%
2.5%
54.8%
2.33
2.62
0.32
1.7%
100
Max Drawdown
-2.6% (2013-12-31..2014-02-01)
-2.1% (2013-04-30..2013-06-29)
95
2013
2014
2015
100
Model Equity %
2015-05-29 = 33.3%
100
75
75
50
50
25
25
2.0
1.5
Stock Contribution
Bond Contribution
1.0
2.0
1.5
1.0
0.5
0.5
0.0
0.0
-0.5
-0.5
-1.0
-1.0
-1.5
-1.5
Monthly Excess Return of Stock/Bond External Model With New Global Market Sentiment Indicator
NSQ2015Q1_21
25
Had the new SHUT indicator been

part of the stock/bond exernal model
(chart at right), performance during the
real-time period would have increased
by 10bps/year over the existing version,
with standard deviation and downside
deviation decreasing by 10 and 20 basis
points, respectively. This alternative
version has an excess returns distribution
similar to the original.
Stock/Bond External Model With New SHUT Indicator Real-Time Performance

Model
Benchmark
125
125
120
120
115
115
110
110
105
105
100
Title
Gain/Annum Std Dev Downside Dev Bat Avg Sharpe Info Ratio Tracking Err
Max Drawdown
Model
10.2%
Benchmark 10.5%
95
4.0%
4.0%
2.4%
2.5%
45.2%
2.51
2.62
-0.25
1.4%
100
-2.0% (2013-04-30..2013-06-29)
-2.1% (2013-04-30..2013-06-29)
95
2013
2014
2015
Replacing both the new global

sentiment and SHUT indicators into the
stock/bond external model (chart below)
led to the best back-tested performance
over the real-time period. This version
picked up 120 bp/year in performance
over the original model, adding 50 bps/
year to its standard deviation of excess
returns and 30 bps/year to downside
deviation. Adding both new indicators
to the model increased the magnitude of
negative excess return in January of 2014;
however, it would have sidestepped the
negative outlier in January 2015.
100
Model Equity %
2015-05-29 = 41.7%
100
75
75
50
50
25
25
Stock Contribution
Bond Contribution
1.0
1.0
0.5
0.5
0.0
0.0
-0.5
-0.5
-1.0
-1.0
-1.5
-1.5
Monthly Excess Return of Stock/Bond External Model With New SHUT Indicator
NSQ2015Q1_22
Stock/Bond External Model With New Global Sentiment and SHUT Indicators Real-Time Performance
130
125
120
130
Model
Benchmark
125
120
115
115
110
110
105
105
100
Title
Gain/Annum
Std Dev
Downside Dev
Bat Avg
Sharpe
Info Ratio
Tracking Err
Max Drawdown
Model
11.2%
Benchmark 10.5%
95
4.6%
4.0%
2.9%
2.5%
51.6%
2.41
2.62
0.43
1.6%
100
-2.2% (2013-12-31..2014-02-01)
-2.1% (2013-04-30..2013-06-29)
95
2013
2014
2015
100
Model Equity %
2015-05-29 = 41.7%
100
75
75
50
50
25
25
1.5
Stock Contribution
Bond Contribution
1.0
Positive Model Returns
1.5
Negative Model Returns
1.0
0.5
0.5
0.0
0.0
-0.5
-0.5
-1.0
-1.0
-1.5
-1.5
Monthly Excess Return of Stock/Bond External Model With New Global Sentiment and SHUT Indicators
NSQ2015Q1_23
26
LISA MICHALSKI, CFA
Associate Director, Custom Research Solutions

Lisa Michalski, CFA, Associate Director, Custom Research Solutions
Department, performs a variety of in-depth analysis, including portfolio
evaluation, equity selection, and asset allocation modeling. Lisa designs
and maintains models, including NDRGs flagship Global Balanced

Account Model and Global Regional Equity Model, for key internal and
external clients. Lisa joined the firm in 2007.
Before joining NDRG, Lisa was a Vice President in Credit Risk Technology
for Goldman Sachs Group, Inc., where she was responsible for quantitative
risk measures of derivative portfolios.
Lisa earned a Master of Science degree in Computational Finance from
the Tepper School of Business, Carnegie Mellon University. She received

her Bachelor of Science degree in Computer Science from the University of
Michigan. Lisa is a CFA charterholder and is a member of the CFA Institute
and CFA Tampa Bay.
MATT BAUER, CFA
Senior Research Analyst
Matt Bauer, CFA, Senior Research Analyst, Custom Research Solutions,
performs a variety of analysis for internal and external clients, including

economic studies, technical indicator testing, custom reporting, and modeling.
Matt joined NDR in 2012.
Prior to joining NDR, Matt worked in a variety of roles at Charles Schwab,
including registered representative, option trader, and most recently as the option
specialist for Schwab Private Client Investment Advisory, where he developed
strategies to generate income and hedge equity risk in client portfolios.
Matt is a graduate of Regis University with a bachelors degree in finance.
Matt is also a CFA charterholder and a member of the CFA Institute, CFA Tampa
Bay, and the Market Technicians Association.
27

sales @ndr.com
www.ndr.com
(800) 241-0621
VENICE
NDRG EDITORIAL BOARD

Ned Davis
Senior Investment Strategist
Lance Stonecypher, CFA

Chief U.S. Equity Sector Strategist
Tim Hayes, CMT

Chief Global Investment Strategist
Ed Clissold, CFA
U.S. Market Strategist
Joseph Kalish
Chief Global Macro Strategist
Brian Sanborn, CFA

Global Quantitative Equity Strategist
600 Bird Bay Drive West

Venice, FL 34285
(941) 412-2300
BOSTON
50 Federal Street
6th Floor
Boston, MA 02110
(617) 279-4860
ATLANTA
DISCLAIMER
The data and analysis contained herein are provided as is and without warranty of any kind, either
expressed or implied. Ned Davis Research, Inc. (NDR), d.b.a. Ned Davis Research Group (NDRG), any NDRG
affiliates or employees, or any third-party data provider, shall not have any liability for any loss sustained
by anyone who has relied on the information contained in any NDRG publication. NDRG disclaims any
2100 RiverEdge Parkway

Suite 750
Atlanta, GA 30328
(770) 308-1128
and all express or implied warranties, including, but not limited to, any warranties of merchantability,
SAN FRANCISCO
not necessarily be updated as views or information change. All opinions expressed herein are
8 Bouverie Street
London
EC4Y 8AX
+44 (0)20 7779 8579
communication reects our analysts opinions as of the date of this communication and will
subject to change without notice. NDRG or its affiliated companies or their respective shareholders,
directors, officers and/or employees, may have long or short positions in the securities discussed
herein and may purchase or sell such securities without notice.
Using any graph, chart, formula or other device to assist in deciding which securities to trade or when to trade
them presents many difficulties and their effectiveness has significant limitations, including that prior patterns
may not repeat themselves continuously or on any particular occasion. In addition, market participants using
such devices can impact the market in a way that changes the effectiveness of such device.
Further distribution prohibited without prior permission. For data vendor disclaimers, refer to
www.ndr.com/vendorinfo.
Copyright 2015 (c) Ned Davis Research, Inc. All rights reserved.
Generate alpha. Identify risk. Choose Ned Davis Research.

Founded in 1980, Ned Davis Research Group is a leading
independent research firm with over 1,100 institutional
clients in over three dozen countries.
With a
range of products and services utilizing a 360

methodology,
we
deliver
Macro
damental
Fun
award-winning
solutions to the worlds leading investment
Idea
management companies. Our clients include

professionals from global investment firms,
banks, insurance companies, mutual funds, hedge
funds, pension and endowment funds, and registered
investment advisors.
Te c h nic a l
S e nt
i m e nt
AC H
LONDON
NDRGs past recommendations and model results are not a guarantee of future results. This
RO
50 California Street
Suite 1500
San Francisco, CA 94111
(415) 277-5477
suitability or fitness for a particular purpose or use.
A
360
PP

NDR

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

NDR

Hochgeladen von

Copyright:

Verfügbare Formate

Ned Davis

NDR Solutions Quarterly

Investment Insight from NDR's Custom Solutions Team

LISA MICHALSKI, CFA Associate Director, Custom Research Solutions

NED DAVIS RESEARCH GROUP

NDR Solutions Quarterly | JUNE 25, 2015

Case Study: Global Balanced Account Model.................................................................................................. 2-3

Real-Time Performance: October 31, 2012 - May 31, 2015........................................................................... 4

Measuring Individual Asset Contribution............................................................................................................ 5

Setting Performance Expectations.......................................................................................................................... 6

Evaluating Real-Time Performance.......................................................................................................................... 7

Model and Sub-Model Decomposition........................................................................................................... 8-11

Indicator Analysis: Stock/Bond Externals.................................................................................................... 12-14

Testing Indicator Alternatives........................................................................................................................... 15-19

Appendix A: Back-Test Supplementary Analysis...................................................................................... 20-22

Appendix B: Real-Time Supplementary Analysis..................................................................................... 23-24

Appendix C: Stock/Bond External Model Alternatives.......................................................................... 25-26

www.ndr.com | Periodical | Issue #CRS201506251

NDR Solutions Quarterly | JUNE 25, 2015

NED DAVIS RESEARCH GROUP

Please see important disclosures at the end of this report.

www.ndr.com | Periodical | Issue #CRS201506251

NDR Solutions Quarterly | JUNE 25, 2015

NED DAVIS RESEARCH GROUP

This paper focuses on attribution analysis for real-time

Performance attribution enables us to determine the

In the pages that follow, we outline an asset allocation

NDR Global Balanced Account Model

CASE STUDY: GLOBAL BALANCED

Benchmark Based On 55% Stocks, 35% Bonds, & 10% Cash

100 Benchmark: MSCI All Country World TR Index

100 Benchmark: Barclays Global Aggregate TR Bond Index

Benchmark: Equal-Weighted 3-month Treasury Composite TR

Please see important disclosures at the end of this report.

All performance numbers do not include transaction costs.

www.ndr.com | Periodical | Issue #CRS201506251

NDR Solutions Quarterly | JUNE 25, 2015

NED DAVIS RESEARCH GROUP

www.ndr.com | Periodical | Issue #CRS201506251

NDR Solutions Quarterly | JUNE 25, 2015

NED DAVIS RESEARCH GROUP

REAL-TIME PERFORMANCE: OCTOBER 31, 2012 MAY 31, 2015

for the model to accumulate substantial gains versus the

NDR Global Balanced Account Model Real-Time Performance

Monthly Data 2012-10-31 to 2015-05-31 (Log Scale)

All performance numbers do not include transaction costs.

Benchmark: MSCI All Country World TR Index

Benchmark: Barclays Global Aggregate TR Bond Index

Benchmark: Equal-Weighted 3-month Treasury Composite TR

Please see important disclosures at the end of this report.

NDR Solutions Quarterly | JUNE 25, 2015

NED DAVIS RESEARCH GROUP

MEASURING INDIVIDUAL ASSET CONTRIBUTION

Weight. If the asset is overweight, the result will be positive; if

Next, we compute the Asset Contribution by multiplying

where r is the periodic excess return of the model over its

Finally, we calculate the overall model excess return by

First, we determine the Overweight vs. Benchmark for each

Model Weight (%)

Asset Return (%)

Model Excess Return (sum):

www.ndr.com | Periodical | Issue #CRS201506251

NDR Solutions Quarterly | JUNE 25, 2015

NED DAVIS RESEARCH GROUP

NED DAVIS RESEARCH GROUP

NED DAVIS RESEARCH GROUP

NED DAVIS RESEARCH GROUP

NED DAVIS RESEARCH GROUP

NED DAVIS RESEARCH GROUP

NED DAVIS RESEARCH GROUP

NED DAVIS RESEARCH GROUP

NED DAVIS RESEARCH GROUP

NED DAVIS RESEARCH GROUP