Maybe It Really Is Different This Time Presentation to: Nomura Quantitative Investment Strategies Conference June 16, 2009 Robert C. Jones [email protected]

Agenda

A

Review of August, 2007

B

The “Stickers” versus the “Adapters”

C

The Stickers’ case: Long-term perspective: HML and WML Normal ebb and flow

D

The Adapters’ case: Quant crowding Adjusting for trading and liquidity

E

The future of quantitative equity management

1

Sudden and dramatic underperformance of previously uncorrelated common quant factors Factor performance in August 2007 scaled to 1% daily volatility1

US: 20%

Revisions

10% 0% -10%

Momentum Value

-20%

Accruals

-30% -40% -50% -60% 1

3

7

9

13

15

17

21

23

27

29

31

August 2007

1

Equivalent to 16% annualized volatility. Scaled to normalize different return series. The graphs above illustrate backtested average daily returns to our target investment portfolios without taking into account transactions costs. These returns are reported for informational purposes only and do not reflect the performance of any GSAM product. These performance results are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. Source: QIS.

2

Sudden and dramatic underperformance of previously uncorrelated common quant factors Factor performance in August 2007 scaled to 1% daily volatility1

Japan: 20% Momentum

10% 0%

Accruals Revisions Value

-10% -20% -30% -40% 1

3

7

9

13

15

17

21

23

27

29

31

August 2007

1

Equivalent to 16% annualized volatility. Scaled to normalize different return series. The graphs above illustrate backtested average daily returns to our target investment portfolios without taking into account transactions costs. These returns are reported for informational purposes only and do not reflect the performance of any GSAM product. These performance results are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. Source: QIS.

3

Sudden and dramatic underperformance of previously uncorrelated common quant factors Factor performance in August 2007 scaled to 1% daily volatility1

UK: 5% Accruals Value Revisions Momentum

0% -5% -10% -15% -20% 1

3

7

9

13

15

17

21

23

28

30

August 2007

1

Equivalent to 16% annualized volatility. Scaled to normalize different return series. The graphs above illustrate backtested average daily returns to our target investment portfolios without taking into account transactions costs. These returns are reported for informational purposes only and do not reflect the performance of any GSAM product. These performance results are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. Source: QIS.

4

Sudden and dramatic underperformance of previously uncorrelated common quant factors Factor performance in August 2007 scaled to 1% daily volatility1

Continental Europe: 5% Momentum Revisions

0% -5%

Value Accruals

-10% -15% -20% -25% 1

3

7

9

13

15

17

21

23

27

29

31

August 2007

1

Equivalent to 16% annualized volatility. Scaled to normalize different return series. The graphs above illustrate backtested average daily returns to our target investment portfolios without taking into account transactions costs. These returns are reported for informational purposes only and do not reflect the performance of any GSAM product. These performance results are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. Source: QIS.

5

The Stickers vs. the Adapters

The Stickers believe this is part of the normal volatility of such strategies •

Long-term perspective: results for HML and WML not outside historical experience.

•

Investors who stick to their process will end up amply rewarded

The Adapters believe that quant crowding has fundamentally changed the nature of these factors •

Likely to be more volatile and offer lower returns going forward

•

Need to adapt your process if you want to add value consistently in the future

Who is right?

Source: Goldman Sachs Asset Management.

6

Historical perspective – HML1 Recent US results for HML do not seem extraordinary relative to historical trends

Cumulative HML daily return, July 1926 – February 2009 450 400

Cumulative return (%)

350 300 250 200 150 100 50 0 -50 2006

2001

1996

1991

1986

1981

1976

1971

1966

1961

1956

1951

1946

1941

1936

1931

1926

-100

Source: Kenneth French Data Library. 1 HML stands for high book-to-price minus low book-to-price as defined by the Kenneth French Data Library. The Fama/French factors are constructed using the 6 valueweight portfolios formed on size and book-to-market. HML is the average return on the two value portfolios minus the average return on the two growth portfolios. http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html. The time period shown above (1926 – 2009) was selected due to data availability.

7

Historical perspective – WML1 Recent US results for WML are also consistent with historical trends

Cumulative WML daily return, January 1927 – February 2009 900 800 Cumulative return (%)

700 600 500 400 300 200 100 0 2007

2002

1997

1992

1987

1982

1977

1972

1967

1962

1957

1952

1947

1942

1937

1932

1927

-100

Source: Kenneth French Data Library. 1 WML stands for winners minus losers as defined by the Kenneth French Data Library. The Fama/French factors are constructed using the 6 value-weight portfolios formed on size and prior (2 -12) returns to construct Momentum. WML is the average return on the two high prior return portfolios minus the average return on the two low prior return portfolios. http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html. The time period shown above (1927 – 2009) was selected due to data availability.

8

Quant crowding

•

Leading up to 2007, quantitative investing had become an increasing percentage of the overall market.

•

When quantitative investors were forced to unwind their positions simultaneously in August 2007, it created the dramatic moves that were synchronized across factors and markets.

•

There is always the potential for more risk and higher correlations for quant factors whenever quant crowding is evident. 1.4% 1.2% 1.0% 0.8% 0.6% 0.4% 0.2%

Dec-08

Mar-08

Jun-07

Sep-06

Dec-05

Mar-05

Jun-04

Sep-03

Dec-02

Mar-02

Jun-01

Sep-00

Dec-99

0.0% Mar-99

Select Quant AUM / Russell 3000 AUM

Total Quant AUM as a percentage of the Russell 3000 Market Cap, March 1999 – December 2008

For illustrative purposes only. The chart shows selected active quantitative manager assets as a percentage of the Russell 3000 Index. The universe of managers examined were those institutions filing holdings data with 13Fs which were picked up in the Thomson ownership database. We screened for quantitative managers who we consider our main competitors based on investment philosophy and other factors we deemed relevant.

9

Historical perspective – Value 5-day standardized returns (Dec-90 to Dec-08)

Looking at intra-month data, results for US Value in August 2007 were well outside of historical norms. 15.0 Standard Deviation

10.0 5.0 0.0 -5.0 -10.0

Dec-08

Dec-07

Dec-06

Dec-05

Dec-04

Dec-03

Dec-02

Dec-01

Dec-00

Dec-99

Dec-98

Dec-97

Dec-96

Dec-95

Dec-94

Dec-93

Dec-92

Dec-91

Dec-90

-15.0

These performance results are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. Please see additional disclosures. Source: Goldman Sachs Asset Management. The OTP is an optimized, unconstrained long-short portfolio which does not take transactions costs into account. It is not meant to reflect performance achieved in any realized accounts, but rather to measure the success of the model's predictions. The returns presented herein are gross and do not reflect the deduction of investment advisory fees, which will reduce returns. These performance results for optimal tilt portfolios are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. With regard to our simulation methodology, we generate risk and return estimates for all assets in our research database using our proprietary model. Simulated performance results do not reflect actual trading and have inherent limitations. Please see additional disclosures. We standardize the 5-day OTP returns by dividing by the rolling 90 day volatility. The time period shown above (1990 – 2008) was selected due to data availability.

10

Historical perspective –Momentum 5-day standardized returns (Dec-90 to Dec-08)

Daily US data for Momentum also shows extreme results 15.0

Standard Deviation

10.0 5.0 0.0 -5.0 -10.0

Dec-08

Dec-07

Dec-06

Dec-05

Dec-04

Dec-03

Dec-02

Dec-01

Dec-00

Dec-99

Dec-98

Dec-97

Dec-96

Dec-95

Dec-94

Dec-93

Dec-92

Dec-91

Dec-90

-15.0

These performance results are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. Please see additional disclosures. Source: Goldman Sachs Asset Management. The OTP is an optimized, unconstrained long-short portfolio which does not take transactions costs into account. It is not meant to reflect performance achieved in any realized accounts, but rather to measure the success of the model's predictions. The returns presented herein are gross and do not reflect the deduction of investment advisory fees, which will reduce returns. These returns are lagged by 20 days for illustrative purposes and are generated by creating an optimal single-factor momentum portfolio and looking at the subsequent returns to the Momentum portfolio 20 days later. We have used returns for the 20 day lagged Momentum portfolio in the above chart to account for short term price reversals. These performance results for optimal tilt portfolios are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. With regard to our simulation methodology, we generate risk and return estimates for all assets in our research database using our proprietary model. Simulated performance results do not reflect actual trading and have inherent limitations. Please see additional disclosures. We standardize the 5-day OTP returns by dividing by the rolling 90 day volatility. The time period shown above (1990 – 2008) was selected due to data availability.

11

A closer look – Value When we look only at liquid stocks, the value effect is more muted and exhibits a flat trend for 20 years

450 400 350 300 250 200 150 100 50 0 -50 -100

Full universe

2006

2001

1996

1991

1986

1981

1976

1971

1966

1961

1956

1951

1946

1941

1936

1931

Large cap universe

1926

Cumulative return (%)

Cumulative daily return of Value; largest US stocks by market capitalization versus full universe, July 1926 – December 2008

Source: Kenneth French Data Library. http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html The large cap universe portfolio shown above is created monthly, where each month the investment universe is reset to the current largest 20% of stocks by market capitalization. The large cap universe portfolio is compared to the full HML universe portfolio, as defined on page 7. The time period shown above (1926 – 2008) was selected due to data availability.

12

A closer look- Momentum When we look only at liquid stocks, the momentum effect is more muted for 20 years

Cumulative daily return of Momentum; largest US stocks by market capitalization versus full universe, July 1926 – December 2008

Cumulative return (%)

900 800

Full universe

700 600

Large cap universe

500 400 300 200 100

2006

2001

1996

1991

1986

1981

1976

1971

1966

1961

1956

1951

1946

1941

1936

1931

1926

0 -100

Source: Kenneth French Data Library. http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html The large cap universe portfolio shown above is created monthly, where each month the investment universe is reset to the current largest 20% of stocks by market capitalization. The large cap universe portfolio is compared to the full WML universe portfolio, as defined on page 8. The time period shown above (1926 – 2008) was selected due to data availability.

13

The impact of transaction costs - Value

When we adjust our Value portfolio to deduct (assumed) trading costs, results look much worse, especially at higher levels of AUM Cumulative daily returns to optimal Value portfolio, February, 1987 – December, 2008 200%

Value portfolio, gross of t-costs Value portfolio, low AUM, net of t-costs

Cumulative return (%)

0%

-200%

-400%

-600%

-800%

Value portfolio, high AUM, net of t-costs 2008

2005

2002

1999

1996

1993

1990

1987

-1000%

Please refer to page 10 for the backtesting methodology used in the Value (B / P) portfolio’s shown above. We define low assets under management (AUM) as an initial portfolio size of $10mm and high AUM as an initial portfolio size of $2bn. These results are gross and net of transactions costs, as defined by our proprietary transaction cost model. The time period shown above (1987 - 2008) was selected due to data availability.

14

The impact of transactions costs - Momentum

When we adjust our Momentum portfolio to deduct (assumed) trading costs, results look much worse, especially at higher levels of AUM Cumulative daily returns to optimal Momentum portfolio, February, 1987 – December, 2008 400%

Momentum portfolio, gross of t-costs Momentum portfolio, low AUM, net of t-costs

200%

Cumulative return (%)

0%

-200%

-400%

-600%

-800%

2008

2005

2002

1999

1996

1993

1990

1987

-1000%

Momentum portfolio, high AUM, net of t-costs

Please refer to page 10 for the backtesting methodology used in the Momentum (Price Momentum: 2 – 12 month return) portfolio’s shown above. We define low AUM (assets under management) as an initial portfolio size of $10mm and high AUM as an initial portfolio size of $2bn. These results are gross and net of transactions costs, as defined by our proprietary transaction cost model. The time period shown above (1987 - 2008) was selected due to data availability.

15

Considering t-costs during optimization - Value

When we add t-costs to the objective function, turnover goes down and net returns improve, but gross returns are much weaker than in the no t-cost case Cumulative daily return to t-cost optimized Value portfolios, February 1987 – December 2008 100%

Cumulative return (%)

Gross of Net of T-cost T-cost High AUM 0.27 0.08 Low AUM 0.70 0.51

80%

Low AUM, gross of t-costs

60%

Low AUM, net of t-costs

40%

High AUM, gross of t-costs

20%

High AUM, net of t-costs 0%

-20%

2008

2005

2002

1999

1996

1993

1987

-40% 1990

Information Ratios Feb 1987 – Dec 2008

Please refer to page 10 for the backtesting methodology used in the Value (B / P) portfolio’s shown above. We define low assets under management (AUM) as an initial portfolio size of $10mm and high AUM as an initial portfolio size of $2bn. These results are gross and net of transactions costs, as defined by our proprietary transaction cost model. The time period shown above (1987 - 2008) was selected due to data availability.

16

Considering t-costs during optimization - Momentum

When we add t-costs to the objective function, turnover goes down and net returns improve, but gross returns are much weaker than in the no t-cost case Cumulative daily return to t-cost optimized Momentum portfolio, February 1987 – December 2008 200%

Low AUM, gross of t-costs

Cumulative return (%)

Gross of Net of T-cost T-cost High AUM 0.73 0.42 Low AUM 1.63 1.34

150%

Low AUM, net of t-costs

100%

High AUM, gross of t-costs High AUM, net of t-costs

50%

0%

2008

2005

2002

1999

1996

1993

1987

-50% 1990

Information Ratios Feb 1987 – Dec 2008

Please refer to page 10 for the backtesting methodology used in the Momentum (Price Momentum: 2 – 12 month return) portfolio’s shown above. We define low AUM (assets under management) as an initial portfolio size of $10mm and high AUM as an initial portfolio size of $2bn. These results are gross and net of transactions costs, as defined by our proprietary transaction cost model. The time period shown above (1987 - 2008) was selected due to data availability.

17

Exposure to current and legacy factors - Value

When we add t-costs to the optimizer, our Value portfolio has exposures to both current and legacy (lagged) factor values Partial correlations of US optimal and lagged Value portfolios, February 1987 – December 2008 1.00

Optimal portfolio, gross of t-costs

0.90

Optimal portfolio net of t-costs at $10 million AUM

0.80

Optimal portfolio net of t-costs at $2 billion AUM

Partial Correlation

0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 -0.10 Lag 1

Lag 20

Lag 60

Lag 120

Source: Goldman Sachs Asset Management. Partial correlation measures the degree of association between the lagged single-factor Value portfolio and the Value optimal tilt portfolio, with the effect of a set of controlling random variables removed. Please refer to page 10 for the backtesting methodology used in the single-factor Value (B / P) portfolio’s shown above. We have shown the above portfolios at various lags (days) to illustrate the effect of factor decay. We define low assets under management (AUM) as an initial portfolio size of $10mm and high AUM as an initial portfolio size of $2bn. These results are gross and net of transactions costs, as defined by our proprietary transaction cost model. The time period shown above (1987 - 2008) was selected due to data availability.

18

Exposure to current and legacy factors - Momentum

When we add t-costs to the optimizer, our Momentum portfolio has exposures to both current and legacy (lagged) factor values Partial correlations of US optimal and lagged Momentum portfolios, February 1987 – December 2008 1.00

Optimal portfolio, gross of t-costs

0.90

Optimal portfolio net of t-costs at $10 million AUM

0.80

Optimal portfolio net of t-costs at $2 billion AUM

Partial Correlation

0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 -0.10 Lag 1

Lag 20

Lag 60

Lag 120

Source: Goldman Sachs Asset Management. Partial correlation measures the degree of association between the lagged single-factor Momentum portfolio and the Momentum optimal tilt portfolio, with the effect of a set of controlling random variables removed. Please refer to page 10 for the backtesting methodology used in the single-factor Momentum (Price Momentum: 2 – 12 month return) portfolio’s shown above. We have shown the above portfolios at various lags (days) to illustrate the effect of factor decay. We define low assets under management (AUM) as an initial portfolio size of $10mm and high AUM as an initial portfolio size of $2bn. These results are gross and net of transactions costs, as defined by our proprietary transaction cost model. The time period shown above (1987 2008) was selected due to data availability.

19

Factor decay - Value

•

The factor decay graphs show how the returns to legacy portfolios decline relative to the returns of current portfolios

•

Recent results show lower means and faster decays than older results, a sign of quant crowding (i.e., faster response to new information) 2.00

1985-1996 1997-2008

Information Ratio

1.50

1.00

0.50

0.00 Lag 1

Lag 5

Lag 20

Lag 60

Lag 120

Source: Goldman Sachs Asset Management. Please refer to page 10 for the backtesting methodology used in the single-factor Value (B / P) portfolio’s shown above. We have shown the above portfolios at various lags (days) to illustrate the effect of factor decay. These results are gross of transaction costs. The time series has been split into two equal time periods to show how factor decay has changed for the more recent time period relative to the historical time period. The time period shown above (1985 - 2008) was selected due to data availability.

20

Factor decay - Momentum

•

The factor decay graphs show how the returns to legacy portfolios decline relative to the returns of current portfolios

•

Recent results show lower means and factor decays than older results, a sign of quant crowding (i.e., faster response to new information) 4.00

1985-1996

3.50

1997-2008

Information Ratio

3.00 2.50 2.00 1.50 1.00 0.50 0.00 Lag 1

Lag 5

Lag 20

Lag 60

Lag 120

Source: Goldman Sachs Asset Management. Please refer to page 10 for the backtesting methodology used in the single-factor Momentum (Price Momentum: 2 – 12 month return) portfolio’s shown above. We have shown the above portfolios at various lags (days) to illustrate the effect of factor decay. These results are gross of transaction costs. The time series has been split into two equal time periods to show how factor decay has changed for the more recent time period relative to the historical time period. The time period shown above (1985 - 2008) was selected due to data availability.

21

Decay rates for other popular factors

Other well-known and popular quant factors have also seen their returns decline and decay rates speed up in recent years Revisions:

Earnings Surprise: 5.00

5.00

1997-2008

4.00

3.00

Information Ratio

Information Ratio

4.00

1985-1996

2.00 1.00 0.00 -1.00

3.00 2.00 1.00 0.00 -1.00

-2.00 Lag 1

Lag 5

Lag 20

Lag 60

-2.00

Lag 120

Lag 1

Lag 5

Lag 20

Lag 60

Lag 120

Reversal:

Accruals: 5.00

7.00 6.00 5.00 Information Ratio

Information Ratio

4.00

3.00

2.00

4.00 3.00 2.00 1.00 0.00 -1.00

1.00

-2.00 -3.00

0.00 Lag 1

Lag 5

Lag 20

Lag 60

Lag 120

Lag 1

Lag 5

Lag 20

Lag 60 Lag 120

Source: Goldman Sachs Asset Management. Please refer to page 10 for the backtesting methodology used in the single-factor portfolio’s shown above. We have shown the above portfolios at various lags (days) to illustrate the effect of factor decay. These results are gross of transaction costs. The time series has been split into two equal time periods to show how factor decay has changed for the more recent time period relative to the historical time period. The time period shown above (1985 - 2008) was selected due to data availability. We define Earnings Surprise as abnormal returns around earnings announcements; Revisions as analysts’ earnings estimate revisions; Accruals as balance sheet accounting accruals; Reversal as one month change in stock price.

22

So now what?

When properly adjusted for liquidity and trading costs, popular quant factors have indeed become less effective recently. The Stickers solution: •

Live with it

•

Sharpe ratios still attractive

The Adapters solution: •

Develop proprietary factors

•

Dynamic allocation to popular factors

•

Incorporate human judgment, but chastened

Source: Goldman Sachs Asset Management.

23

Sources of alpha

•

Alpha can only be derived when prices don’t accurately reflect all public information (i.e., price ≠ fair value).

•

This can only happen when investors over- or under-react to information.

•

Thus, the Value effect is really an over-reaction effect; Momentum is an under-reaction effect.

•

Investors tend over-react to information that confirms their prior beliefs and under-react to information that contradicts those beliefs.

•

We can develop proprietary factors by looking for other situations where investors systematically over- or under-react to information.

Source: Goldman Sachs Asset Management.

24

Factor results in August 2007 Scaled to 1% daily volatility1

US: 10% 0% -10% -20% -30% -40% -50% -60% 1

3

7

9

13

15

17

21

23

27

29

31

August 2007

Mostly proprietary factors

Proprietary factors

More well-known factors

1

Equivalent to 16% annualized volatility. Scaled to normalize different return series. The graphs above illustrate backtested average daily returns to our target investment portfolios without taking into account transactions costs. These returns are reported for informational purposes only and do not reflect the performance of any GSAM product. These performance results are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. The factor categorizations shown above were determined based on what QIS perceives to be the level of crowding and popularity of quantitative factors. The most common factors are included in the “more well-known” category, while the most unique factors included in the “mostly proprietary factors” category. Source: QIS

25

Factor results in August 2007 Scaled to 1% daily volatility1

Japan: 10% 0% -10% -20% -30% -40% -50% 1

3

7

9

13

15

17

21

23

27

29

31

August 2007

Mostly proprietary factors

Proprietary factors

More well-known factors

1

Equivalent to 16% annualized volatility. Scaled to normalize different return series. The graphs above illustrate backtested average daily returns to our target investment portfolios without taking into account transactions costs. These returns are reported for informational purposes only and do not reflect the performance of any GSAM product. These performance results are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. The factor categorizations shown above were determined based on what QIS perceives to be the level of crowding and popularity of quantitative factors. The most common factors are included in the “more well-known” category, while the most unique factors included in the “mostly proprietary factors” category. Source: QIS

26

Factor results in August 2007 Scaled to 1% daily volatility1

UK: 10% 5% 0% -5% -10% -15% -20% -25% -30% 1

3

7

9

13

15

17

21

23

28

30

August 2007

Mostly proprietary factors

Proprietary factors

More well-known factors

1

Equivalent to 16% annualized volatility. Scaled to normalize different return series. The graphs above illustrate backtested average daily returns to our target investment portfolios without taking into account transactions costs. These returns are reported for informational purposes only and do not reflect the performance of any GSAM product. These performance results are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. The factor categorizations shown above were determined based on what QIS perceives to be the level of crowding and popularity of quantitative factors. The most common factors are included in the “more well-known” category, while the most unique factors included in the “mostly proprietary factors” category. Source: QIS

27

Factor results in August 2007 Scaled to 1% daily volatility1

Continental Europe: 10% 5% 0% -5% -10% -15% -20% -25% -30% -35% 1

3

7

9

13

15

17

21

23

27

29

31

August 2007

Mostly proprietary factors

Proprietary factors

More well-known factors

1

Equivalent to 16% annualized volatility. Scaled to normalize different return series. The graphs above illustrate backtested average daily returns to our target investment portfolios without taking into account transactions costs. These returns are reported for informational purposes only and do not reflect the performance of any GSAM product. These performance results are backtested based on an analysis of past market data with the benefit of hindsight, do not reflect the performance of any GSAM product and are being shown for informational purposes only. The factor categorizations shown above were determined based on what QIS perceives to be the level of crowding and popularity of quantitative factors. The most common factors are included in the “more well-known” category, while the most unique factors included in the “mostly proprietary factors” category. Source: QIS

28

Decay rates for popular factors

An equal-weighted portfolio of Value and Momentum has experienced lower means and accelerated decay rates in recent times

4.00

1985-1996 1997-2008

3.50

Information Ratio

3.00 2.50 2.00 1.50 1.00 0.50 0.00 Lag 1

Lag 5

Lag 20

Lag 60 Lag 120

Source: Goldman Sachs Asset Management. Please refer to page 10 for the backtesting methodology used in the equal-weighted Value and Momentum factor (Value =B / P, Momentum = 2 – 12 month return) portfolio’s shown above. We have shown the above portfolios at various lags (days) to illustrate the effect of factor decay. These results are gross of transaction costs. The time series has been split into two equal time periods to show how factor decay has changed for the more recent time period relative to the historical time period. The time period shown above (1985 - 2008) was selected due to data availability.

29

Decay rates for proprietary factors

Information Ratio

However, our proprietary factors have not seen the same decline in mean and/or acceleration in decay rates in recent times

4.00

1985-1996

3.50

1997-2008

3.00 2.50 2.00 1.50 1.00 0.50 0.00 Lag 1

Lag 5

Lag 20

Lag 60 Lag 120

Source: Goldman Sachs Asset Management. Please refer to page 10 for the backtesting methodology used in the composite weighted proprietary factor portfolio’s shown above. The proprietary factor composite was determined based on what QIS perceives to be the level of crowding and popularity of quantitative factors. We have shown the above portfolios at various lags (days) to illustrate the effect of factor decay. These results are gross of transaction costs. The time series has been split into two equal time periods to show how factor decay has changed for the more recent time period relative to the historical time period. The time period shown above (1985 - 2008) was selected due to data availability.

30

Factor timing

•

Factors become vulnerable (riskier) when they become “too popular”

•

By monitoring factor popularity, we can adjust exposures to reflect changes in risk or expected return

Cumulative performance of intra-industry Value strategy in the US, January 1928 – February 2009 450

Above average: 5.9 Below average: 3.6

350

1.2

Cumulative return (%)

400

1.1 1.0

300 250

0.9

200 0.8

150 100

0.7

50

0.6

0

Value spread

2008

2003

1998

1993

1988

1983

1978

1973

1968

1963

1958

1953

1948

1943

1938

1933

0.5 1928

-50

1 The

QIS value spread

Subsequent 12-month return (%) to HML when Value spread is:

Return to value

QIS value spread is a metric formulated by the QIS team to capture the industry-adjusted dispersion in company over/under valuations. The above strategy was formed based on industry-adjusted B / P metrics. The value strategy is a portfolio that overweights those equities with high B / P (undervalued firms) and underweights those equities with low B / P. We then calculated the return to the portfolio and repeated the process on a monthly frequency. Simulated performance results do not reflect actual trading and have inherent limitations. Source of return data: Kenneth French Data Library. http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html. Source of firm book data (value spread): Kenneth French, Compustat. Due to data availability, we used Kenneth French Data Library for return to value and value spread data prior to 1978 (annual availability). Subsequently, we used Compustat for value spread data (monthly availability) and Kenneth French Data Library for return data after 1978. Due to the different nature of the value spread data before and after 1978, we indexed data in the latter period to have the same starting point as in the former. This information discusses general market activity, industry or sector trends, or other broad-based economic, market or political conditions and should not be construed as research or investment advice. Please see additional disclosures. 31

Fix the underlying problem

•

Quants have embraced objective models to forecast returns because they do not trust biased human judgment

•

An alternative solution is to train analysts and traders to recognize and control their biases

•

Not strictly a quant approach, but uses insights gleamed from quantitative analysis

•

Can also use quant/statistical models to improve analysts forecasts of the underlying fundamentals (EPS, ROE, growth, payout, etc.)

Source: Goldman Sachs Asset Management.

32

Conclusions

•

Historical results for many popular quant factors look much weaker when we adjust for liquidity and trading costs

•

Crowding has fundamentally changed the outlook for many of these factors: ⎯ Likely to prove more volatile and provide lower Sharpe ratios in the future ⎯ Stronger results still possible for smaller managers who focus in less liquid stocks or markets

•

Some opportunities for quant managers ⎯ Develop proprietary factors that capitalize on over- and underreaction ⎯ Dynamic allocation to popular factors based on perceived crowding ⎯ Work with traditional analysts and traders to minimize over- and under-reaction

Source: Goldman Sachs Asset Management.

33

General notes

Tracking Error (TE) is one possible measurement of the dispersion of a portfolio’s returns from its stated benchmark. More specifically, it is the standard deviation of such excess returns. TE figures are representations of statistical expectations falling within “normal” distributions of return patterns. Normal statistical distributions of returns suggests that approximately two thirds of the time the annual gross returns of the accounts will lie in a range equal to the benchmark return plus or minus the TE if the market behaves in a manner suggested by historical returns. Targeted TE therefore applies statistical probabilities (and the language of uncertainty) and so cannot be predictive of actual results. In addition, past tracking error is not indicative of future TE and there can be no assurance that the TE actually reflected in your accounts will be at levels either specified in the investment objectives or suggested by our forecasts. References to indices, benchmarks or other measures of relative market performance over a specified period of time are provided for your information only and do not imply that the portfolio will achieve similar results. The index composition may not reflect the manner in which a portfolio is constructed. While an adviser seeks to design a portfolio which reflects appropriate risk and return features, portfolio characteristics may deviate from those of the benchmark. The strategy may include the use of derivatives. Derivatives often involve a high degree of financial risk because a relatively small movement in the price of the underlying security or benchmark may result in a disproportionately large movement in the price of the derivative and are not suitable for all investors. No representation regarding the suitability of these instruments and strategies for a particular investor is made. Past performance is not indicative of future results, which may vary. The value of investments and the income derived from investments can go down as well as up. Future returns are not guaranteed, and a loss of principal may occur. Backtested Performance Backtested performance results do not represent the results of actual trading using client assets. They do not reflect the reinvestment of dividends, the deduction of any fees, commissions or any other expenses a client would have to pay. If GSAM had managed your account during the period, it is highly improbable that your account would have been managed in a similar fashion due to differences in economic and market conditions. Simulated Performance Simulated performance is hypothetical and may not take into account material economic and market factors that would impact the adviser’s decision-making. Simulated results are achieved by retroactively applying a model with the benefit of hindsight. The results reflect the reinvestment of dividends and other earnings, but do not reflect fees, transaction costs, and other expenses, which would reduce returns. Actual results will vary. Expected Returns Expected return models apply statistical methods and a series of fixed assumptions to derive estimates of hypothetical average asset class performance. Reasonable people may disagree about the appropriate statistical model and assumptions. These models have limitations, as the assumptions may not be consensus views, or the model may not be updated to reflect current economic or market conditions. These models should not be relied upon to make predictions of actual future account performance. GSAM has no obligation to provide updates or changes to such data. Opinions expressed are current opinions as of the date appearing in this material only. No part of this material may, without GSAM’s prior written consent, be (i) copied, photocopied or duplicated in any form, by any means, or (ii) distributed to any person that is not an employee, officer, director, or authorised agent of the recipient. Copyright © 2009, Goldman, Sachs & Co. All rights reserved. Rev#21110.OTHER.OTU

34