高盛Goldman Sachs自上而下选股框架--宏观和微观相结合120611

2012年6月11日

宏观和微观相结合：高盛自上

而下选股框架

证券研究报告将宏观观点转化为选股建议的工具 我们推出了一套基于宏观因素描绘、微观层面比较以及对商业周期各阶段投资分析的自上而下选股框架，从而将宏观观点与微观投资建议相关联。 宏观和微观因素对于回报都非常重要 对于亚洲市场上市值占比60%左右的股票而言，宏观因素（如全球经济增速、本地经济增速和金融状况）都是重要的回报推动因素。此外，估值、每股盈利市场共识预测调整以及技术性指标等微观因素也与随后3个月的回报高度相关。 宏观 + 微观 = 超额收益 对于全球领先指标(GLI)中定义的四阶段商业周期，我们基于各阶段的宏观和微观特征对股票进行筛选。回溯测试显示，采用这种方法获得的回报有望超过传统买入/持有策略带来的回报。 经济衰退阶段的选股：降低贝塔值和对增长的敏感性；买入受益于政策放松的股票 我们的GLI 指标显示，自4月份以来我们已进入经济衰退阶段。在此背景下，我们特别关注于受宏观面影响和微观面支撑的程度处于适当水平的股票：NAB 、Orica 、恒安国际、中石油、腾讯控股、长江实业、恒生银行、HDFC 、ITC 、BRI 、LS Corp 、Shinhan 、CIMB 、SM Inv.、JCC 、Delta 、TSMC 和BK Bank 。我们将定期更新这些股票建议。 对部分市场的部分板块采用自上而下的选股框架 资料来源：FactSet 、MSCI 、CEIC 、高盛全球经济商品和策略研究 刘劲津, CFA

+852-2978-1224 https://www.360docs.net/doc/5a14224531.html,u@https://www.360docs.net/doc/5a14224531.html,

高盛（亚洲）有限责任公司

慕天辉, CFA

+852-2978-1328 timothy.moe@https://www.360docs.net/doc/5a14224531.html,

高盛（亚洲）有限责任公司

Caesar Maasry

+852-2978-7213 caesar.maasry@https://www.360docs.net/doc/5a14224531.html,

高盛（亚洲）有限责任公司

鄧啟志

+852-2978-0722 richard.tang@https://www.360docs.net/doc/5a14224531.html, 高盛（亚洲）有限责任公司

Sunil Koul

+852-2978-0924 sunil.koul@https://www.360docs.net/doc/5a14224531.html,

高盛（亚洲）有限责任公司

高盛与其研究报告所分析的企业存在业务关系，并且继续寻求发展这些关系。因此，投资者应当考虑到本公司可能存在可能影响本报告客观性的利益冲突，不应视本报告为作出投资决策的唯一因素。有关分析师的申明和其他重要信息，见信息披露附录，或参阅https://www.360docs.net/doc/5a14224531.html,/research/hedge.html 。由非美国附属公司聘用的分析师不是美国FINRA 的注册/合格研究分析师。本报告仅供分发给高盛机构客户。

Introducing GS top-down stock picking framework

In this report, we introduce a framework to help connect the macro environment to

single stock selection.

Specifically, we believe this framework would be helpful to:

?Formalize our existing top-down stock picking logic and approach in a

disciplined and statistically-tested manner that is also intuitively

appealing

?Better comprehend how macro factors influence individual stock returns

and help connect macro trends to actionable stock ideas

?Expand our implementation focus to include more single stock ideas as

well as theme baskets and derivative overlays

?Complement our sector analysts’ views and our bottom-up stock

selection processes such as Asia-Pacific Conviction Lists and GS Sustain, which

focus principally on operating returns and industry position

?Provide a tool which can add further perspective to investors’ own stock

selection processes

The exhibit below summarizes the key building blocks and logic flow in this report.

The key building blocks and approach of our stock-selection framework

Source: FactSet, I/B/E/S, Goldman Sachs Global ECS Research estimates.

Executive summary

Top-down and bottom-up are the two prevalent approaches to securities analysis. However,

neither is perfect— macro-focused investors may overlook micro factors and industry dynamics

while stock pickers may sometimes miss the big picture.

With an objective of combining the two approaches and translating macro views into actionable

stock ideas (and, hopefully, good returns), we introduce a top-down stock selection framework

which builds upon macro factor profiling, micro specific comparisons, and business-

cycle-based investing.

Key conclusions and investment implications in this report are:

1.Macro analysis is important, even to stock pickers. Our regression model shows

select macroeconomic factors are statistically significant return drivers (R2>40%) for

around 250 stocks in Asia, representing 58% of MXAPJ market cap. This

underscores the importance of macro analysis even to bottom-up-

oriented investors. Macro analysis is particularly effective in Hong Kong, China,

Singapore, and sectors including energy, financials, and materials.

2. Some macro variables are more important than others. Investors can be

overwhelmed by macro data, but many macroeconomic variables are highly correlated.

Stock pickers can simplify the process of macro monitoring by focusing on six factors,

namely, market risk, local growth, policy/liquidity, CPI, oil prices, and global

growth.

3.Micro specifics have linked well with ensuing returns. Empirically, micro

considerations such as valuation, consensus EPS and target price changes, and

technical entry levels have shown very strong relationships with ensuing 3-month

returns. These micro considerations complement the macro perspective and address

the extent to which fundamentals are discounted in share prices.

4.Our two-tier stock selection strategy may help performance. We have

established a trading algorithm based on our global leading indicator and its derived

business cycle— expansion, slowdown, contraction, and recovery— to

implement our macro and micro analysis. Backtests of our two-tier strategy

suggest performance can be enhanced to a meaningful extent by

considering these macro and micro factors.

5. Useful tool. This framework is flexible and can complement the investment process for

different types of investors.

Caveats: A tool, not a cure-all

We believe this approach adds value, but we also recognize its limitations.

Regression models are static and assume mean-reversion, and need to be

updated and refined to adapt to changing fundamentals.

Stock ideas: Investing in a contraction phase

The latest GLI reading suggests the global economy has entered a contraction phase. Against

this backdrop, we would:

a. Reduce market risk “beta”;

b. Buy stocks with low sensitivity to local growth;

c. Accumulate policy easing beneficiaries;

d. Own stocks that may outperform in a disinflationary environment;

e. Buy stocks that may benefit from lower oil prices;

f. Overweight stocks that are less sensitive to global growth momentum.

Stocks which have these macro characteristics and have shown favorable micro readings are

shown in Exhibit 1. Exhibit 1: We like these stocks because of their favorable macro exposure and compelling micro profile relative to their regional peers

Stock recommendations for June 2012 (Priced as of June 5)

Note (1): These stocks are rated Buy or Neutral by Goldman Sachs Research except SM Investments which is NC. We use consensus estimates for SM Investments. Note (2): * denotes the stock is on our regional Conviction List. “Tick” indicates stock that ranks top-30 percentile within each factor relative to its market peers and they perform well in our specified macro environment. Revisions and sentiment are based on forward 12-month consensus EPS.

Source: Factset, I/B/E/S, Goldman Sachs Global ECS Research estimates.

Key exhibits for chart lovers

We highlight the important exhibits (takeaways) as follows:

? Exhibit 7: Where top-down analysis may be more applicable in terms of markets and

sectors.

? Exhibit 8: Stocks that are the most and least sensitive to different macro factors across

markets.

? Exhibit 17: The empirical relationships between micro parameters and ensuing returns.

? Exhibits 28 to 38: Market summary pages which detail the factor loading for each stock

under our study universe.

Part 1: Mapping stock returns to macro exposure

Defining our study universe

To ensure the practicability of the regression results, we choose to focus on MXAPJ constituents

with at least US$1bn of index market cap, over US$5mn of average daily value traded (ADVT) in

the past 6 months, and at least 5 years of listing history. There are 412 stocks in MXAPJ

which fit these requirements (as of April 30, 2012) and they represent 86% of

MXAPJ index market cap . These stocks are mostly located in Australia, China, Korea

and India , with a sector concentration in financials, IT and telecoms (Exhibits 2 and 3). Exhibit 2: Liquid, large-cap stocks are mostly found in Australia, China, Korea and India Free-float market cap distribution by markets

Exhibit 3: ...and they are concentrated in sectors including financials, IT, and telecoms Free-float market cap distribution by sectors

Source: FactSet, MSCI, Goldman Sachs Global ECS Research. Source: FactSet, MSCI, Goldman Sachs Global ECS Research.

Defining the dependent variables (returns)

We choose 3-month price returns as the dependent variable in our regression model as we

attempt to strike a balance between high frequency trading and the “buy-and-hold” approach.

Returns are quoted in local currency , on our assumption that foreign investors should treat

the currency decisions separately from the stock selection decisions and the impact of FX

changes could be partly reflected/captured in the underlying macro trends (e.g. exports in CAI

and FCI).

Choosing and testing independent variables (macro factors)

While there are a large number of macro variables that equity investors could focus on, we elect

to limit our independent variables to 14 macro factors which theoretically should drive stock prices.

In other words, this is not an exhaustive list of macro variables which might influence stock

returns but what we have found to be generally influential at a market level. See Asia Pacific:

Portfolio Strategy: What macro indicators matter for Asian markets?, May 7, for details.

We also test the statistical significance of

2nd derivatives and lead/lags for each of the factors

in our multi-factor regression model to ensure statistically important relationships will be

accounted for.

Given many of these variables are inter-correlated and are essentially linked to similar sets of

fundamental drivers, we run a correlation matrix to eliminate those with strong directional

relationships to improve the ensuing regression results (Appendix 1). In cases where macro

factors are strongly correlated among each other within a category (e.g. growth), we prefer our

Australia 20%China 23%Hong Kong 8%India 11%Indonesia

2%Korea 16%Malaysia 2%

Philippines 0%

Singapore 5%Taiwan 9%Thailand 4%

Banks

23%

Information

Technology 14%

Telecom

11%

Materials

10%

Energy

9%Industrials

7%

Cons Disc

7%

Cons Stap

6%

Property

6%Insurance and other financial services 4%

Utilities 2%

Health Care

1%

proprietary indexes such as Current Activity Index (CAI), Financial Conditions Index (FCI),

and Global Leading Indicator (GLI) given their broader representation and statistically-tested

robustness (see Exhibit 4).

Following the above steps, we conclude that a significant part of an individual stock’s return

variations can be reasonably explained by 6 broad macro measurements: Market risk (MXAPJ),

domestic growth (CAI), domestic liquidity/policy (FCI), domestic inflation (CPI), oil

prices (WTI), and global growth (GLI).

The final step is to establish linkages between returns and these macro factors by using

regression techniques, namely simple linear and multiple factor regression models .

Recognizing the advantages and deficiencies of these modeling techniques, we choose to base

our stock selection analysis on the former and the market analysis on the latter.

? Simple linear regression by a single factor (e.g. six separate regression

models) allows us to estimate the factor loadings for the independent

variables without running into multicollinearity issues. This approach

may work better in capturing the maximum total exposure to a single

factor. That said, it doesn't account for the impact of other significant

variables and therefore is not robust to changes in the relationship

between macro factors over time. Even if we run a number of single

factor regressions, the resulting individual R2s from this are not additive

and we cannot statistically prove the explanatory power of each factor

for returns.

? Multiple factor regression with stepwise elimination helps form a quantifiable

relationship (equation) as to what factors are important and to what extent they, when all

treated as independent variables, have historically affected stock returns (when one

factor changes and others are held constant). The drawback is that the (high) correlation

among macro variables lowers the precision of the regression estimates, which is likely

to lead to estimates being very inaccurate for some stocks when the regressions are

carried out across a large universe of single stocks. If we subsequently use the analysis

to pick stocks with the highest/lowest sensitivities there is a risk that we will also end up

maximizing exposure to estimation errors. Exhibit 4: The independent variables in our regression model are representative of the principal macro categories that tend to influence stock returns

Macro variables and the inputs to our regression model

Note: We use the next 3 month yoy growth for CAI (local/US/EU) as they show significantly higher correlations with returns, which provides better indicative power on

returns and fits our purpose of mapping returns to macro exposure better. See Appendix 1 for details. Source: FactSet, I/B/E/S, Goldman Sachs Global ECS Research estimates.

Input

specifications Chosen

variables

Regional index returns Log P(t)‐Log P(t ‐3)

Country index returns Log P(t)‐Log P(t ‐3)Regional sector index returns Log P(t)‐Log P(t ‐3)

Local IP Log (Avg. of last 3 monthly yoy growth)

Local export growth Log (Avg. of last 3 monthly yoy growth)Local retail sales

Log (Avg. of last 3 monthly yoy growth)Local CAI Log (Avg. of next 3 monthly yoy growth)Local PMI Log (Avg. of latest 3 mom growth)Local CPI Log (Avg. of last 3 monthly yoy growth)Local CPI Local FCI Log (Avg. of last 3 monthly yoy growth)Local FCI

WTI Price Log WTI(t)‐Log WTI(t ‐3)US CAI Log (Avg. of next 3 monthly yoy growth)EU CAI Log (Avg. of next 3 monthly yoy growth)

EM GDP Log (Avg. of last 3 monthly yoy growth)GLI momentum Log (latest 3 mom changes)MXAPJ returns

Local CAI

WTI

GLI

I n

d e p e

n d

e

n t

v a r

i a b l

e s

Stock beta Domestic growth Domestic liquidity/policy Global growth

Exhibit 5: Our model shows that stock returns can be reasonably explained by 6

market/macro factors

Components our multi-factor regression model

Intercept other factors Source: Goldman Sachs Global ECS Research.

Regression results (1): Understanding where macro analysis may apply

Top-down approach works for some stocks, but not all. Our model has yielded reasonably high R2 (over 40%) for 247 stocks out of the total 412 sample universe (58%

of market cap). It also means the remaining 165 stocks could be more sensitive to micro

factors as opposed to macro forces if one takes R2 of 40% as the threshold1.

From a market standpoint, the top-down approach seems to work better for Hong Kong, China and Singapore, while the bottom-up study appears more suitable for Australia and the ASEAN-4 markets. We think these results reflect the following:

a. Hong Kong, China (HK-listed) and Singapore have open economies and

free capital markets, meaning that the local stock markets are more sensitive

to global macro forces and capital flows relative to the region.

b. The dynamics between domestic and externally-driven demand for Australia

can differ—currently, the domestic economy is hampered by the (until recently)

strong currency, weak property market and issues in the banking system while

commodity exports remain resilient due to demand from global EMs. This

dichotomy makes a static top-down analysis less effective.

c. The domestic demand component and generally low foreign investors

participation in the ASEAN-4 have resulted in lower returns volatility for these

smaller markets. As such, the top-down framework has to be adjusted by

local factors to make it more applicable and effective to explain returns.

By sector, the average R2 is generally low for defensives including telcos, utilities and healthcare stocks, suggesting: a) their share prices are not sensitive to macro

changes, relative to the aggregate market; b), their price sensitivity to the market risk

factor is low as reflected by the regression results; and, c) a micro-focused approach is

required to generate alpha in these sectors.

At the other end of the spectrum, a top-down approach could be effective for energy,

financials, and materials given the high R2, which conceptually makes sense as these sectors are closely linked with global dynamics via real demand and financial channels.

1 Note that these R2s are derived from our regression models and the macro factors may not represent the true macro profile for our study universe.

Exhibit 6: Macro factors appear important return drivers (R2 more than 40%) for 60% of the

stocks and 58% of market cap in MXAPJ Accumulated distribution of R 2 based on our study universe

Source: Goldman Sachs Global ECS Research estimates.

Exhibit 7: Top-down analysis works for select economic groups but not all

Average R2 based on our multi-factor regression

Source: FactSet, I/B/E/S, Goldman Sachs Global ECS Research estimates.

Regression results (2): Making sense of it

We ran macro-factor regressions for more than 400 stocks in Asia and the results are organized

by their factor loadings in each market in Exhibit 8 (we show the top-3 stocks under each market

factor only; regression results for the full universe are shown in the country summary pages).

0%10%20%30%

40%

50%60%70%80%

90%

100%0%10%20%30%40%50%60%70%80%90%100%Top ‐down approach may work

better

Macro factors explain >40% of return variations

%of market cap

R2

Bottom ‐up approach may work

better

Macro factors explain<40% of return variations

Bottom 25 percentile Top 25 percentile

Exhibit 8: We group stocks under different macro buckets based on their factor loadings

Stocks with highest/lowest factor loadings by market

Note (*): Highest (positive) coefficients for FCI means stocks have historically performed better relative to their benchmark when financial conditions tighten. Note (1): We rank stocks by their factor loadings based on the output from our simple linear regression model. Multiple factor regression models using specific independent variables are required to estimate the relative significance of variables and their explanatory power on returns.

Note (2): We only show stocks which rank in the top-80 percentile in terms of R2 ranking for that particular factor.

Note (3): We exclude Philippines because only 4 stocks satisfy our liquidity requirements.

Source: Goldman Sachs Global ECS Research estimates.

To better interpret and make sense of the regression output, we think it is useful to look at a few

examples below.

Example 1: Samsung Electronics (005930 KR)

The regression R2 is low for Samsung Electronics (33%), suggesting the stock’s returns

could be more sensitive to specific micro factors than the high-level macro variables that

we test.

?While Samsung is perceived by investors as a DM growth proxy, its share does not

appear to be well linked with global growth momentum (GLI is eliminated in the stepwise

regression). This coincides with the stock’s substantial outperformance in the past few

years due to its improved competiveness and strong product cycle (e.g. Galaxy

smartphones).

?It is positively correlated with domestic growth and liquidity conditions, which seems

normal, although their statistical significance is lower than the market risk factor.

Exhibit 9: The low R2 for Samsung suggests that micro factors could be more important in driving the stock price Stepwise regression results (Samsung Electronics)

Source: FactSet, CEIC, MSCI, Goldman Sachs Global ECS Research estimates.

Example 2: ICBC (1398 HK)

?Our model shows 74% of ICBC’s share price variations can be explained by the four

factors in the stepwise regression model.

?ICBC’s sensitivity to the market risk factor is high (high beta), partly reflecting the

relatively high growth and policy volatility (or market concerns) in China.

?It is positively correlated with inflation and policy easing as the bank may benefit from

loan pricing and higher loan quota when these macro conditions move in its favor.

?The share prices tend to outperform (underperform) the aggregate market when global

growth decelerates (accelerates), as the stock is not as sensitive as the MXAPJ

aggregate to global growth momentum.

Exhibit 10: Policy easing should bode well for ICBC but its GDP growth exposure may not be as high as many have perceived

Stepwise regression results (ICBC)

Source: FactSet, CEIC, MSCI, Goldman Sachs Global ECS Research estimates.

Example 3: WesFarmers (WES AU)

? The regression R2 is high compared with other Australian stocks in our study sample,

meaning that top-down approach may make more sense for WES relative to other

Australian stocks.

? The stock has showed low sensitivity to MXAPJ returns in local currency terms.

? Higher CPI tends to bode well for stock returns, as high inflation may allow the company

to raise prices more easily.

? While the local exposure of its business operations may suggest high sensitivity to

domestic growth activities, domestic CAI is omitted from the regression. This may reflect

the hybrid structure of the Australian economy as discussed on page 7. activities link well with export growth cycles, adding complications when interpreting regression results

Stepwise regression results (WesFarmers)

Source: FactSet, CEIC, MSCI, Goldman Sachs Global ECS Research estimates.

# of obs.60R2

74%

# of obs.142R2 50%

Example 4: TATA Consultancy Services (TCS IN)

? It has one of the highest R2 (74%) among the Indian stocks in the MXAPJ universe

(average 46%).

? The stock can be classified as a low-beta stock given its low return coefficient with

MXAPJ (c.0.67).

? The stock has tended to move very closely with GLI momentum (1% of GLI momentum

change has historically led to 15% of share price movement), consistent with TCS’s

business concentration in global developed markets. Exhibit 12: TATA consultancy’s share prices seem very sensitive to global growth given the company’s geographic exposure

Stepwise regression results (TATA Consultancy Services)

Source: FactSet, CEIC, MSCI, Goldman Sachs Global ECS Research estimates.

# of obs.86R2 74%

Part 2: Adding micro overlays—valuation, micro fundamentals and technical indicators—to enhance returns

Part 1 aims to match individual equities’ returns to select macro factors. However, the return

attribution exercise is insufficient to build a sensible trading strategy on its own, because:

a. No market or stock in our sample universe has close to a 100% fit (R2) in our regression

model, suggesting some unobserved or uncaptured variables, which we believe are

mostly micro-related, are also significant performance drivers;

b. Stock returns are essentially a joint function of earnings growth and valuation changes.

Our analysis helps partly explain returns variations using macro factors but does not

take into account the returns that macro drivers have on risk premia and therefore the

willingness of investors to pay for equities for a given level of earnings.

c. In some cases, share prices may have already reflected stock-specific themes/exposure;

and pricing signals (especially shorter-term ones) are difficult to capture in macro-factor

regression models.

As such, we incorporate valuation parameters, micro fundamental variables, and various

technical indicators into our broader framework. We believe these additional considerations

could help us form an objective view on stocks’ micro profile and help answer the questions of

“how much is priced in” and “entry level”, which have shown strong relationships with price

returns (more on this later).

Valuation parameters

We focus on the stock’s current valuations to gauge how the stock’s fundamentals are being

priced by the market and to avoid buying/selling at full/depressed valuation levels.

Specifically, taking the conclusions from our recent work: Asia Pacific: Portfolio Strategy: Global

Strategy Paper: No. 3: AsiaPac Valuation: What works, and when, March 12, 2012, we have

chosen forward P/E, trailing D/Y, and trailing P/B, which are proven to be significant returns

indicators (higher significance to medium-term returns) according to our analysis, as the key

metrics.

While we have also proved that P/CF has strong predictive power on forward returns at a market

level, we deliberately exclude this in our analysis given its short and unstable time series at an

individual stock level.

Micro fundamentals

While we have established relationships between macro factors and stock returns in Part 1,

stocks react to macro forces because they tend to impact earnings or earnings expectations,

which can be quantified and captured by the changes of consensus EPS and expected returns

(target price). In this vein, we look at following indicators to assess stock’s micro dynamics using

consensus data:

?Earnings revision (magnitude)—percentage of month-on-month forward 12-month

EPS changes;

?Earnings sentiment (breadth)—percentage of net earnings upgrades/downgrades

versus the total number of consensus estimates (mom);

?Consensus target price—percentage changes (mom).

Technical indicators

Broadly speaking, technical indicators can be grouped into four categories—momentum,

volatility, trend and volume . Given the numerous forms of technical indicators, we focus on

those that are most commonly used and are comparable from a time series standpoint. The key

objective here is to form a view on short-term entry levels.

1. RSI (momentum): We use 14-day RSI to gauge the near-term price momentum of

particular stocks. We prefer 14-day over longer-dated RSIs given its higher volatility

relative to longer-dated RSIs, which better fits our objective of evaluating entry point for

shorter-term investing.

2. Bollinger bands (volatility): For computational purpose, we look at %b , as opposed

to the actual moving averages and the two bands (upper and lower) to quantify where

the current price is relative to its recent range in standard deviation terms. Low %b

indicates current prices are closer to the lower Bollinger band and vice versa.

3. Moving Average Convergence/Divergence (MACD) (trend): We take the

differentials between MACD line (12 day-26 day exponential moving average (EMA))

and MACD signal line (9-day EMA), which is the histogram on a typical MACD chart, to

gauge the short-term trend and price momentum of the stock. High positive values

reflect strong price momentum in recent trading periods and vice versa.

4. Volume: Most volume-based indicators require a time-series perspective to form trading

signals; hence, an absolute number often does not tell us much. Volume signals can

vary so decision-making is not stable. As such, we exclude volume-related

indicators in our micro score calculation .

Exhibit 13: We use these micro parameters to form a view on the stocks’ fundamentals, aiming to enhance the

risk/reward of our top-down stock recommendations

A summary of micro parameters

Source: FactSet, I/B/E/S, Goldman Sachs Global ECS Research estimates.

The effectiveness of micro overlays—an empirical study

While these micro parameters are commonly regarded as important elements to potential returns,

their actual implications and contribution to returns are unclear to us up to this point.

As such, we test the conditions and formats under which these micro factors matter most to

potential returns, and design our trading strategies accordingly. Input specifications Forward P/E P(t)/fEPS(t)Trailing P/B P(t)/tBPS(t)

Trailing D/Y P(t)/tDPS(t)

EPS revisions fEPS(t)/fEPS(t ‐1)

‐1

EPS sentiment {#up rev(t)‐#down rev(t)}/#estimate(t)Consensus target price TP(t)/TP(t ‐1)‐1

Momentum 14‐day RSI

Volatility Bollinger Bands (%b)

Trend MACD line ‐ MACD signal line Composite micro score

M i

c r

o p

a

r a

m e

t e r s Valuations Micro fundamentals Technical indicators

First, we aggregate and organize ex-ante, subsequent 3-month stock returns based on the nominal values of these parameters. The sample size is statistically significant as there are more than 50,000 datapoints for each parameter, given our study universe contains over 400 stocks with more than 12 years of monthly history (i.e. 400*144). The results support the conventional wisdom of buying stocks at low valuations and technically sound entry levels, and when consensus expectations rise, ensuing returns tend to be strong and vice versa. Details are shown in Appendix 2.

We then standardize all the parameters based on their z-scores to ensure comparability and compatibility of the dataset, and to allow us to form an objective view of the stock’s micro attractiveness with existing (ex-ante) data-points. Key points to note:

?Forms of standardization: Time series and cross-sectional analysis are commonly used data standardization methods. While they have their own analytical advantages

and drawbacks, we choose the forms under which the parameters have historically

generated more differentiated (higher or lower) returns. In this vein, we take time

series z-scores for valuations and micro revision data, and cross-sectional z-

scores for technical indicators.

?Valuations—Extreme valuations usually lead to significant subsequent returns: In our recent research, Global Strategy Paper: No. 3 - AsiaPac Valuation:

What works, and when, we found that the levels of valuation, on a standalone basis,

have low correlation with ensuing returns over a relatively short time horizon (3-6 months) at the regional/market level. However, the picture looks slightly different at the stock

level. We note that when prevailing stock valuations are close to 1.5 to 2 standard

deviations (s.d.) to the attractive side (3-year rolling z-scores), subsequent 3-month

performance tends to be strong (Exhibit 14). The opposite is not very obvious, unless

when valuations are extremely demanding (2 s.d. to the unattractive side).

?Micro fundamentals—Consensus view changes are a reasonably good indicator of short-term returns: Unsurprisingly, upgrades of consensus earnings

(both magnitude and breadth) and/or target prices usually lead to favorable price returns.

However, it is noteworthy that we are comparing ex-ante earnings and target price

changes with subsequent 3-month returns, meaning that: a) observable consensus view

changes do drive actual stock returns; and, b) it may take some time for the market to

discount the incremental new consensus expectations (Exhibit 15).

?Technical indicators—The trend is your friend: Exhibit 16 shows, fairly consistently, stocks with high technical scores (low RSI, MACD, %b compared with

peers) tend to outperform those that embrace demanding technical entry levels.

However, the extent to which stocks have historically outperformed/ underperformed is

lower on our technical scores (vs. valuation and micro scores), reflecting that valuations

and fundamentals have comparatively higher contributions to return variations than

technical indicators.

?Composite micro score—an objective view on stock’s micro attractiveness: While the each of the three categories of micro parameters appear to be significant

determinant of short-term returns on its own, we believe a combined measurement

(simple average of the three z-scores) could be even more helpful for investors to gauge

risk/reward because: a) it gives a comprehensive and objective assessment of stocks’

micro profile using observable market data, without involving stocks’ specific operating

and industry-wide expectations; b) it gives higher alpha (positive and negative) than the

individual scores may reflect, meaning that it is probably a better variable to consider in

our stock-picking framework (Exhibit 17).

Exhibit 14: Levels of valuations seem to have strong

impact on returns, especially when valuations are at

extremes 3-month price returns (loc) vs. valuation z-scores

Exhibit 15: Positive earnings revisions (in terms of both magnitude and breadth) and target price upgrades do bode well for performance 3-month price returns (loc) vs. micro fundamentals z-scores Note: High z-scores mean lower fP/E, tP/B and high tD/Y. Note: High z-scores mean positive EPS revisions, EPS sentiment, and consensus target price upgrades. Source: FactSet, Goldman Sachs Global ECS Research.

Source: FactSet, Goldman Sachs Global ECS Research.

Exhibit 16: Technical factors are helpful to explain

returns variations, although not as much as valuations

and fundamentals

3-month price returns (loc) vs. technical z-scores

Exhibit 17: Our composite z-scores provide an objective assessment on stocks’ micro profile and reasonably strong indications to forward returns 3-month price returns (loc) vs. composite micro z -cores Note: High z -scores mean low RSI, MACD and %b.

Source: Bloomberg, Goldman Sachs Global ECS Research.

Source: FactSet, Bloomberg, Goldman Sachs Global ECS Research.

2.7%

6.6% 6.6%

4.8% 4.6%

4.2%

5.6%

8.1%10.1%

0%

2%

4%6%8%10%12%

14%

16%-2-1.5-1-0.500.51 1.52

Valuation score at different entry levels (z score)

Average returns in

our study

universe

Avg. subsequent 3m returns

2.7% 5.9%

3.2% 3.7%

4.2% 6.7%7.1%7.5%9.9%0%2%4%6%8%10%12%14%16%-2-1.5-1-0.500.51 1.52Micro fundamentals score at different entry levels (z score)Avg. subsequent 3m returns Average returns in our study universe 1.6%

2.8%

3.6%

4.2%

4.9% 6.4% 6.4% 6.8%7

.1%

0%

2%4%6%8%10%

12%

14%

16%

-2-1.5-1-0.500.51 1.52

Technical score at different entry levels (z score)

Average returns in

our study universe Avg. subsequent 3m returns 2.0% 3.9% 2.9% 4.5% 4.8% 5.7%8.2%9.5%15.2%0%2%4%6%8%10%12%14%16%-2-1.5-1-0.500.51 1.52Composite micro score at different entry levels (z score)Avg. subsequent 3m returns Average returns in our study universe

Part 3: Back-testing our strategies—Cycle-based trading algorithms may help performance

Building trading algorithms around the GLI

Parts 1 and 2 form the core analytical foundations for our stock selection process. The next step

revolves around designing a trading algorithm and testing (and refining) its effectiveness.

Leveraging the work by our global team in its recent paper: Global Economics Paper No: 214,

Acceleration Matters: Asset Returns and the Business Cycle, May 16, 2012, we define the

macro business cycle using our proprietary Global Leading Indicator (GLI), which

also serves as a signal for our trading strategies.

Simply put, our global team uses the interaction of GLI growth (mom) with GLI acceleration

(changes of mom growth) to define four phases of the business cycle (Exhibit 18):

1. Expansion: Positive growth and positive acceleration.

2. Slowdown: Positive growth and negative acceleration.

3. Contraction: Negative growth and negative acceleration.

4. Recovery: Negative growth and positive acceleration.

In each of the four phases, we examine past market returns (MXAPJ) and historical trends of

macro variables in our regression model to understand how investors should position in the

different periods of the economic cycle. Key insights (and our decision rules) are:

?In an expansion phase, investors should turn aggressive by going long stock beta

and growth proxies (including oil-related exposure). Policy tends to stay neutral

during this phase given the lagged effect of growth on inflation.

?During an economic slowdown, equity returns may stay positive although they are less

obvious than in the expansion phase. Investors should focus on late-cycle plays

including inflation beneficiaries and commodity stocks as growth/inflation

tradeoff deteriorates. Policy tends to become tighter, so stocks that are sensitive

(insensitive) to liquidity may underperform (outperform).

?Defensive is the core theme during economic contraction periods as high

beta equities and growth proxies are likely to be under pressure.

Policymakers tend to loosen monetary policy as growth is challenged, and stocks with

favorable exposure to liquidity should trade well relative to the benchmark.

?Easing starts to take effect and growth begins to accelerate into the recovery stage.

Equities returns are mixed as valuation compression tends to overpower the impact

of nascent growth and earnings upgrades. In other words, investors should own growth

proxies but not necessarily overweight beta. Financial conditions tend to stay very easy

and asset/rate-sensitive slices may outperform.

Exhibit 18: GLI mom growth and acceleration define the four phases of the business cycle and our stock selection decision process

GS GLI momentum and our trading algorithms

Note: “1” means owning the top-40 percentile (by factor sensitivity) of stocks under each market factor and owning the bottom-40 for “-1”. “0” means owning 30 to 70 percentile of the distribution.

Source: Goldman Sachs Global ECS Research estimates.

Back-testing our trading strategy

We test this strategy based on the following procedures:

a. Create a benchmark portfolio which includes top-80 percentile of stocks by their R2

ranking in each of the six macro factors in order to include stocks to which the top-down

approach may apply. We also believe this is a better performance proxy than MXAPJ

because of survivorship bias in our study universe.

b. Take the latest GLI reading as input and allocate macro exposure (1, 0 or -1) based on

the trading algorithm we defined in Exhibit 18. Specifically, “1” means owning the top-40

percentile of stock (based on their factor loadings) under that particular factor for

each market, “-1” means buying the bottom-40 percentile, and “0” refers to the 30th to

70th percentile of the distribution.

c. We then rank stocks by their macro attractiveness, as defined by the simple

average of their ranking in all 6 macro factors (depending on the business cycle), to filter stocks with reasonably compelling macro exposure (i.e. these stocks may not score well in all six macro filters but they have relatively high ranking in all categories on average.)

This forms our “macro-only” portfolio.

d. From (c), we select stocks in each market with the highest aggregate micro z-scores

(top-15 percentile) and form our “macro + micro” portfolio. The portfolio is rebalanced on a quarterly basis and the price returns are measured in local currency.

Exhibit 19: The logic flow/mechanism of our top-down selection framework

Source: Goldman Sachs Global ECS Research estimates.

Evaluating the back-test results: It is an alpha, not beta strategy ?Performance: Over the past ten years, our “macro + micro” portfolio has gained 326%, versus benchmark of 165% on a market-cap-weighted basis, translating into 11.4pp

average annualized outperformance according to our backtest (as of May 31). The

portfolio has generated accumulated price return of 1914% since 2002 on an equal-

weighted basis, outperforming the benchmark (672%) by 1232pp. This translates into

88pp average outperformance per annum versus the benchmark (Exhibit 20). “Macro”

and “macro + micro” portfolios outperformed the benchmark 29 and 24 out of 42 quarters since 2002 respectively.

?Representation: Our portfolio has consisted of at least 23 stocks across the full study period, representing around 5% of the universe by number of stocks and 5% by free-

float market cap. Portfolio constituents are proportionately distributed across markets

according to their representation in the universe as designed by our constraints. 295

stocks have been included in our portfolio at least once.

?Volatility: Realized volatility (annualized) of our portfolio has been tracking in line with the benchmark except during 1H09 and 1H12 when the overall market volatility was high.

Given the size (number of stocks in the portfolio) of the portfolio, we consider its realized

volatility as reasonable.

?Alpha or beta?: As shown in Exhibit 23, both portfolios have outperformed the benchmark (on average) in all the economic phases since 2002 on an equal-weighted

basis, suggesting that: a) our strategies are not entirely driven by beta exposure; b)

Some elements of alpha are embedded in our portfolios as they have outperformed in

both expansion and contraction phases2.

?Macro vs. micro: Interestingly, the returns differentials between our “macro only” and “macro + micro” portfolios reflect that one strategy might outperform the other under

different market conditions. For example, “macro” fared better than “macro + micro”

during the bull market from 2005 to 2007 but underperformed in 2009, and we believe

this can be explained by the micro filters that we have put in place to screen out high-

valuation and momentum stocks during that period3. In the case of 2009, given many

stocks were trading at undemanding valuations and technical levels post the Global

Financial Crisis, our micro filters were generally not binding constraints (Exhibit 24). This

shows that investors may be better off if they can relax their valuation/micro

standards at the onset of a market uptrend; however, micro disciplines still

add alpha over time.

Caveats: Risks and limitations

1. Our backtest is conducted on an in-sample basis (i.e. regression for factor loading and

backtest start at the same time), a less preferred approach to out-of-sample test from a

statistical standpoint. However, we are constrained by data availability which is prevalent for Asian stocks where listing history is generally short.

2. Historical GLI readings are subject to revision risk. As such, the indication of economic

turning points by the GLI is more accurate on an ex-post basis.

3. We have not assumed any trading and transaction costs in our backtest returns

calculations. Realized returns could be meaningfully different from the results.

4.Our trading algorithms could be subject to data-mining risk (e.g. we favor certain macro

exposures in different business-cycle phases based on historical pattern).

2 We use monthly rebalancing returns to calculate our portfolios’ performance in different business-cycle phases to better

capture the changes in GLI-derived trading signals.

3 We take the simple average of the valuation, micro fundamentals, and technical scores. Investors may adjust the weighting

of each category depending on market conditions.