Lecture 2 – Active Investment I

AFIN3052 APPLIED PORTFOLIO MANAGEMENT

Fundamentals: Performance Measurement

This class

© Lasse H. Pedersen
Overview of Performance Measurement
• Alpha and Beta
• Risk–Reward Ratios
• Estimating Performance Measures
• Time Horizons and Annualizing Performance Measures
• High Water Mark
• Drawdown
• Adjusting Performance Measures for Illiquidity and Stale Prices
• Performance Attribution
• Back-tests and Track Records

© Lasse H. Pedersen
Alpha and Beta
• Excess return 𝑅𝑅𝑡𝑡𝑒𝑒 = 𝑅𝑅𝑡𝑡 − 𝑅𝑅 𝑓𝑓 separated into alpha and beta:

𝑅𝑅𝑡𝑡𝑒𝑒 = 𝛼𝛼 + 𝛽𝛽𝑅𝑅𝑡𝑡𝑀𝑀,𝑒𝑒 + 𝜀𝜀𝑡𝑡

• β = market risk, tendency to follow the market
• α = excess return, adjusted for market risk, due to HF trading skills (or luck)
• εt = idiosyncratic risk

• Market neutral hedge fund: β = 0

market neutral excess return = 𝑅𝑅𝑡𝑡𝑒𝑒 − 𝛽𝛽𝑅𝑅𝑡𝑡𝑀𝑀,𝑒𝑒 = 𝛼𝛼 + 𝜀𝜀𝑡𝑡

𝐸𝐸 market neutral excess return = 𝛼𝛼

© Lasse H. Pedersen
Why Separate Alpha and Beta?

𝑅𝑅𝑡𝑡𝑒𝑒 = 𝛼𝛼 + 𝛽𝛽𝑅𝑅𝑡𝑡𝑀𝑀,𝑒𝑒 + 𝜀𝜀𝑡𝑡

• β RM,e: can achieved using index fund
• No need to pay large fees for this (e.g. Vanguard will do it for around 0.1%)
• Beta necessary for risk management:
― If you mix a HF into a portfolio with other market risk, the beta risk will not go away
• How large fee should investors at most pay?

© Lasse H. Pedersen
Adjusting for More than the Market

• Three-factor model of Fama and French (1993):

𝑅𝑅𝑡𝑡𝑒𝑒 = 𝛼𝛼 + 𝛽𝛽𝑀𝑀 𝑅𝑅𝑡𝑡𝑀𝑀,𝑒𝑒 + 𝛽𝛽𝐻𝐻𝐻𝐻𝐻𝐻 𝑅𝑅𝑡𝑡𝐻𝐻𝐻𝐻𝐻𝐻 + 𝛽𝛽 𝑆𝑆𝑆𝑆𝑆𝑆 𝑅𝑅𝑡𝑡𝑆𝑆𝑆𝑆𝑆𝑆 + 𝜀𝜀𝑡𝑡

• High minus low (HML): Value strategy based on B/M
• Small minus big (SMB): Size bet based on market capitalization
• What does alpha mean here?

• Alpha and betas with respect to all your current risk factors are relevant

© Lasse H. Pedersen
Risk-Reward Ratios

• The reward for investments is

― expected return, or
― expected abnormal return, that is, alpha
• Positive reward is good and negative reward is bad
― but is higher expected return always better?
• What about risk?
• Returns/alphas depend on how a strategy is scaled
― a twice-leveraged strategy has twice the return/alpha of an unleveraged version of the same
strategy
• Solution:
― We need to consider reward-to-risk ratios

© Lasse H. Pedersen
Sharpe Ratio

𝐸𝐸 𝑅𝑅 − 𝑅𝑅 𝑓𝑓
𝑆𝑆𝑆𝑆 =
𝜎𝜎 𝑅𝑅 − 𝑅𝑅 𝑓𝑓
• SR measures the “reward” for taking risk, per unit of risk that you take
• Reward = expected return over the risk free rate
• Risk = standard deviation of return

© Lasse H. Pedersen 8
Information Ratio
𝐸𝐸 𝑅𝑅 − 𝑅𝑅𝑏𝑏
𝐼𝐼𝐼𝐼 =
𝜎𝜎 𝑅𝑅 − 𝑅𝑅𝑏𝑏
• Where Rb is the return on a benchmark b (everything usually annualized, see later)
• The denominator is also called the “tracking error”
• Sometimes, the benchmark exposure is estimated using a regression

𝑅𝑅𝑡𝑡𝑒𝑒 = 𝛼𝛼 + 𝛽𝛽𝑅𝑅𝑡𝑡𝑏𝑏,𝑒𝑒 + 𝜀𝜀𝑡𝑡

• (The first definition implicitly sets β = 1.) In the general case, the IR is defined as
𝛼𝛼
𝐼𝐼𝐼𝐼 =
𝜎𝜎 𝜀𝜀
• NB: Some HF managers use a cash benchmark (Rb =0) and compute IR simply as
𝐸𝐸 𝑅𝑅
𝐼𝐼𝐼𝐼 =
𝜎𝜎 𝑅𝑅

© Lasse H. Pedersen
Alpha-to-Margin Ratio
– You can’t eat risk-adjusted returns.
• Whether you can or not, depend on how much leverage you can apply
• AM ratio: The return on a maximally leveraged version of a long/short strategy:

𝛼𝛼
𝐴𝐴𝐴𝐴 =
margin
• Example: margin requirement = 10% means that you can apply 10:1 leverage
• AM ratio depends on
― The reward per unit of risk, namely the Information ratio
― The risk you can apply per unit of margin equity:

𝜎𝜎(𝜀𝜀)
𝐴𝐴𝐴𝐴 = 𝐼𝐼𝐼𝐼 ×
margin
• Compare: ROE from corporate finance

© Lasse H. Pedersen
Adjusting the Risk Measure
• Risk-adjusted return on capital (RAROC):
𝐸𝐸 𝑅𝑅 − 𝑅𝑅 𝑓𝑓
𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 =
economic capital
• Economic capital: amount you need to set aside to sustain worst-case losses on the strategy with a
certain confidence, e.g. VaR.

• Sortino ratio (S):

𝐸𝐸 𝑅𝑅 − 𝑅𝑅 𝑓𝑓
𝑆𝑆 = 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
𝜎𝜎
• The downside risk (or downside deviation) is calculated as the standard deviation of returns truncated
above some minimum acceptable return (MAR):

𝜎𝜎 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 = 𝜎𝜎 𝑅𝑅 1 𝑅𝑅<𝑀𝑀𝑀𝑀𝑀𝑀

© Lasse H. Pedersen
Estimating Performance Measures
• Past average performance:
geometric average = 1 + 𝑅𝑅1 × 1 + 𝑅𝑅2 × ⋯ × 1 + 𝑅𝑅𝑇𝑇 1/𝑇𝑇 −1
1
arithmetic average = 𝑅𝑅� = 𝑅𝑅1 + 𝑅𝑅2 + ⋯ + 𝑅𝑅𝑇𝑇
𝑇𝑇

• Risk:
1
variance estimate = R1 − 𝑅𝑅� 2 + R2 − 𝑅𝑅� 2 + ⋯ + RT − 𝑅𝑅� 2
𝑇𝑇−1
standard deviation estimate = variance estimate

© Lasse H. Pedersen
Annualizing Performance Measures
• Arithmetic returns with n measurement periods per year:
ERannual = ER × n
• Geometric returns:
ERannual = (1 + ER)n – 1
• Variance:
varannual = var × n
• Standard deviation/ volatility:
σannual = σ × 𝑛𝑛
• Sharpe ratio:
SRannual = ERannual / σannual = SR × 𝑛𝑛

© Lasse H. Pedersen
Monitoring P&L in Real Time: The Effect of Time Horizon
Measurement horizon Sharpe ratio Loss probability
Four years 2 2.3%
Year 1 16%
Quarter 0.5 31%
Month 0.3 39%
Trading day 0.06 47.5%
Minute 0.003 49.9%

 Measured Sharpe ratio depends on horizon

SR = SRannual / 𝑛𝑛

 Probability of losing depends on SR:

𝐸𝐸 𝑅𝑅𝑒𝑒 = 𝜎𝜎 𝑆𝑆𝑆𝑆
Pr 𝑅𝑅𝑒𝑒 < 0 = Pr 𝐸𝐸 𝑅𝑅𝑒𝑒 + 𝜎𝜎𝜎𝜎 < 0 = Pr(𝑁𝑁 < −𝑆𝑆𝑆𝑆)

© Lasse H. Pedersen
High Water Mark
• Consider a hedge fund’s price of shares or its cumulative return Pt , where
Pt = Pt-1 × (1 + Rt)
• The high water mark (HWM) is the highest price Pt (or highest cumulative return) it has achieved in the
past:

HWMt = maxs≤t Ps

• Often hedge funds only charge performance fees when their returns are above their HWM.
• Hence, if they have experienced losses, they must first make these back and only charge performance
fees on the profits above their HWM.

© Lasse H. Pedersen
Drawdown
• An important risk measure for a hedge fund strategy is its drawdown (DD). The drawdown is the
cumulative loss since losses started.
DDt = (HWMt – Pt) / HWMt
• Experiencing large drawdowns is costly and risky:
― can lead to redemptions from investors
― concerns from counterparties, for example, prime brokers increasing margin requirements or
completely pulling the financing of the hedge fund’s positions.
• When evaluating a strategy, people sometimes consider its maximum drawdown (MDD) over some
past time period:
MDDT = maxt≤T DDt
• If the cumulative returns are arithmetic, Pt = Pt-1 + Rt, then the drawdown is also defined in an additive
way,
DDt = HWMt – Pt.

© Lasse H. Pedersen
High Water Mark and Drawdown: Example

© Lasse H. Pedersen
Adjusting Performance Measures for Illiquidity and Stale Prices

Example
• Suppose “Late Capital Management” (LCM) invests 100% in the market
• LCM always marks to market 1 month late
• What is 𝑐𝑐𝑐𝑐𝑐𝑐 𝑅𝑅𝑡𝑡𝐿𝐿𝐿𝐿𝐿𝐿 , 𝑅𝑅𝑡𝑡𝑀𝑀 ?
• What are the estimated values of α and β in:

𝑅𝑅𝑡𝑡𝐿𝐿𝐿𝐿𝐿𝐿,𝑒𝑒 = 𝛼𝛼 + 𝛽𝛽𝑅𝑅𝑡𝑡𝑀𝑀,𝑒𝑒 + 𝜀𝜀𝑡𝑡

• Is investing in LCM a good hedge against the market?

© Lasse H. Pedersen
Why the Example May be More Real than You Think
• Illiquid securities often do not trade
• Hence, month end price is a “stale price”
• No public price for OTC securities
• Hence, prices do not reflect all the volatility of the market
• Co-movement with the market mis-measured
• “Marketing supportive accounting”

© Lasse H. Pedersen 19
Do Hedge Funds Hedge?
• Many hedge funds claim to “hedge,” more precisely, to be market neutral
• This is helpful because
― It provides diversification for investors
― Excess returns over Rf are value added (not just compensation for systematic risk)
― Justifies high fees
• Do hedge funds have an incentive to hedge perfectly, i.e. choose β = 0 ?

© Lasse H. Pedersen 20
Adjusting Performance Measures for Illiquidity and Stale Prices
• Simple method: use longer-term data (e.g., quarterly instead of monthly)
• More sophisticated method: include past market returns in the regression:
𝑅𝑅𝑒𝑒 = 𝛼𝛼 adjusted + 𝛽𝛽0 𝑅𝑅 𝑀𝑀,𝑒𝑒 + 𝛽𝛽1 𝑅𝑅 𝑀𝑀,𝑒𝑒 + . . . +𝛽𝛽𝐿𝐿 𝑅𝑅 𝑀𝑀,𝑒𝑒 + 𝜀𝜀
𝑡𝑡 𝑡𝑡 𝑡𝑡−1 𝑡𝑡−𝐿𝐿 𝑡𝑡
• The alpha in this multivariate regression accounts for stale market exposure
― captures the hedge fund’s value added beyond its exposure to both current and past market
moves
• We can then estimate the “true” all-in beta as

𝛽𝛽all−in = 𝛽𝛽0 + 𝛽𝛽1 + . . . +𝛽𝛽𝐿𝐿

• What are 𝛽𝛽all−in and 𝛼𝛼 adjusted for LCM in the example above?
• Adjusted Information ratio based on the all-in hedged returns, 𝑅𝑅𝑡𝑡𝑒𝑒 − 𝛽𝛽all−in𝑅𝑅𝑡𝑡𝑀𝑀,𝑒𝑒 :

𝐸𝐸 𝑅𝑅𝑡𝑡𝑒𝑒 − 𝛽𝛽all−in𝑅𝑅𝑡𝑡𝑀𝑀,𝑒𝑒 𝛼𝛼 adjusted

𝐼𝐼𝑅𝑅 adjusted = =
𝜎𝜎 𝑅𝑅𝑡𝑡𝑒𝑒 − 𝛽𝛽all−in 𝑅𝑅𝑡𝑡𝑀𝑀,𝑒𝑒 𝜎𝜎(𝜀𝜀)

21
One Beta Versus Many

Evidence from Asness, Krail, and Liew (2001)

22
Sharpe Ratios Three Ways
unhedged seemingly really
hedged hedged
𝑅𝑅𝑡𝑡𝑒𝑒 𝑅𝑅𝑡𝑡𝑒𝑒 − 𝛽𝛽 𝑅𝑅𝑡𝑡𝑀𝑀,𝑒𝑒 𝑅𝑅𝑡𝑡𝑒𝑒 − 𝛽𝛽all−in 𝑅𝑅𝑡𝑡𝑀𝑀,𝑒𝑒

Evidence from Asness, Krail, and Liew (2001) © Lasse H. Pedersen

Performance Attribution
• Hedge funds frequently review what factors are driving their returns, a process called performance
attribution.
• Look back over the previous quarter, say, and review
― which trades were the main positive return contributors and which ones detracted
• This is useful both for the hedge fund’s
― communication with its clients
 provides insight into the investment process
 the drivers of returns
 the risk factors to which they are exposed.
― internal planning and evaluations
 which investment strategies appear to be working
 which traders tend to make successful investments.

© Lasse H. Pedersen
Performance Measures and Track Records
• Should a performance measure be:
― Adjusted for trading costs?
― Before and after management fees?
• What is important for
― Investors?
― HF managers / management company?
• “Realized” performance measures:
― Risk and expected return estimated using realized returns
― E.g. rolling 12 month SR
• Track record = realized performance measure, after trading costs and after fees

© Lasse H. Pedersen
Fundamentals: Finding Alpha

This class
Sources of Alpha
• We are interested in strategies that can be expected to continue to make
money
― a repeatable process that generates alpha
• To find such a repeatable alpha process, one must understand the economics
hiding behind the profits.
― For every buyer, there is a seller
• If you don't know who the sucker is, it's you
Information
• Markets must be in an equilibrium level of inefficiency (Grossman-Stiglitz (1980)):
― reflecting enough information to make it difficult to make money
― but not so efficient that no one wants to collect information and trade on it
• Production of information, e.g.
― Fundamental analysis of companies and their future profit prospects
― Giving management ideas on how to improve a company or cut costs
― Serving on boards and creditor committees
― Discovering frauds
• Access to information
― Do NOT do insider trading!
• News that travel slowly into prices
― Post–earnings-announcement drift
― Initial underreaction and delayed overreaction
― Trends and momentum
Compensation for Market Liquidity Risk
• Trader Talk: They’ll let you in, but they won’t let you out.
• Market liquidity risk: risk that you cannot get out
• Investors want to be compensated for taking market liquidity risk
― Therefore, illiquid securities are cheap and earn higher average gross returns.
― For instance, convertible bonds tend to be cheap relative to their theoretical value
• Liquidity risk a reason why the standard capital asset pricing model (CAPM) does not work well in
practice.
― Investors care about a security i’s return Ri net of its transaction cost TCi
• Liquidity-adjusted CAPM (Acharya and Pedersen (2005)):

𝑖𝑖 𝑀𝑀 𝑖𝑖 ,𝑇𝑇𝑇𝑇 𝑀𝑀 𝑖𝑖 ,𝑅𝑅 𝑀𝑀 𝑖𝑖 𝑀𝑀
𝐸𝐸 𝑅𝑅𝑖𝑖 = 𝑅𝑅 𝑓𝑓 + 𝐸𝐸 𝑇𝑇𝐶𝐶 𝑖𝑖 + 𝛽𝛽𝑅𝑅 ,𝑅𝑅 + 𝛽𝛽𝑇𝑇𝑇𝑇 − 𝛽𝛽𝑇𝑇𝑇𝑇 − 𝛽𝛽 𝑅𝑅 ,𝑇𝑇𝑇𝑇 𝜆𝜆

29
Compensation for Market Liquidity Risk
• Liquidity-adjusted CAPM (Acharya and Pedersen (2005)):

𝑖𝑖 𝑀𝑀 𝑖𝑖 ,𝑇𝑇𝑇𝑇 𝑀𝑀 𝑖𝑖 ,𝑅𝑅 𝑀𝑀 𝑖𝑖 𝑀𝑀
𝐸𝐸 𝑅𝑅𝑖𝑖 = 𝑅𝑅 𝑓𝑓 + 𝐸𝐸 𝑇𝑇𝐶𝐶 𝑖𝑖 + 𝛽𝛽𝑅𝑅 ,𝑅𝑅 + 𝛽𝛽𝑇𝑇𝑇𝑇 − 𝛽𝛽𝑇𝑇𝑇𝑇 − 𝛽𝛽 𝑅𝑅 ,𝑇𝑇𝑇𝑇 𝜆𝜆

• 𝐸𝐸 𝑇𝑇𝐶𝐶 𝑖𝑖 : expected transaction costs

• λ : risk premium
𝑖𝑖 𝑀𝑀
• 𝛽𝛽𝑅𝑅 ,𝑅𝑅 : the standard market beta
𝑖𝑖 ,𝑇𝑇𝑇𝑇 𝑀𝑀
• 𝛽𝛽𝑇𝑇𝑇𝑇 : premium for commonality in liquidity
𝑖𝑖 ,𝑅𝑅 𝑀𝑀
• 𝛽𝛽𝑇𝑇𝑇𝑇 : premium for securities that becomes illiquid when the market falls
𝑖𝑖 𝑀𝑀
• 𝛽𝛽𝑅𝑅 ,𝑇𝑇𝑇𝑇 : premium for securities that drop in value during a liquidity crisis
Liquidity Crisis

• The liquidity-adjusted CAPM also shows what happens during a liquidity crisis:
― Transaction costs increase
― The required return increases as investors need even higher compensation for market liquidity risk
― As a result, prices drop sharply
Trading on Market Liquidity Risk
• A hedge fund can trade illiquid securities and earn the market liquidity risk premium, if
― Low trading cost
― Long holding period
• Examples:
― Convertible bond arbitrage
― Fixed income arbitrage
• Making markets:
― order flow is fragmented and price bounces around the fundamental value
― Market makers take advantage of this, i.e. provide a liquidity service
― Take the other side of these trades and smoothing out the price fluctuations
― The compensation for the risks associated with this liquidity service is the profit due to bid–ask
spreads or market impacts
• Examples:
― High-frequency hedge funds often effectively play the role of market makers
― Some hedge funds are effectively market makers in OTC markets
Compensation for Funding Liquidity Risk
• Funding liquidity risk :
― Risk that a hedge fund cannot fund the position throughout the life of the trade
― Risk of being forced to unwind positions as the fund hits/nears a margin constraint
• Some securities are difficult to finance
― High margin requirements, mi
• Capital-constrained investors want compensation for tying up capital, resulting in a margin CAPM
(Gârleanu and Pedersen (2009)):
𝐸𝐸 𝑅𝑅𝑖𝑖 = 𝑅𝑅 𝑓𝑓 + 𝛽𝛽𝑖𝑖 𝜆𝜆 + 𝑚𝑚𝑖𝑖 𝜓𝜓
• 𝜓𝜓 : margin premium
• Said differerently: Securities have the same alpha-to-margin ratio, AMi

𝐴𝐴𝑀𝑀𝑖𝑖 = 𝜓𝜓

33
Compensation for Market and Funding Liquidity Risk

• Market and funding liquidity risks operate at the same time:

market liquidity risk funding liquidity risk

𝐸𝐸 𝑅𝑅𝑖𝑖 = 𝑅𝑅 𝑓𝑓 + 𝛽𝛽𝑖𝑖 𝜆𝜆 + +
compensation compensation

• In fact, they re-enforce each other through

― Liquidity spirals
Event Risk
• Another type of risk compensation hidden behind hedge fund returns is a type of insurance premium
• Insurance companies:
― profit from selling protection against your house burning down
• Merger arbitrage hedge funds
― profit from selling insurance against a merger deal failing.
• The same type of insurance economics underlie several other event-driven strategies
Demand Pressure
• Sometimes a security faces unusual buying pressure, while other securities may be abandoned
― Limited arbitrage means that these demand pressures affects prices
― A contrarian trading strategy becomes profitable
• Hedging, e.g.:
― large hedging demand for buying index options, especially put options
• Institutional frictions, e.g.:
― Certain investors cannot hold non–investment-grade bonds
― Passive investors in commodities and the S&P GSCI index roll
• Behavioral biases, e.g.:
― Some securities may capture investors’ imaginations and create a demand for their shares, while
others may go overlooked
― For instance, companies that interact directly with many customers may face an excess demand
for their shares, such as Krispy Kreme, internet stocks
Liquidity Risk and Demand Pressure: Price Level and Fluctuations
• Two effects of liquidity risk: price level and fluctuations

Fundamental value if liquidity is perfect

Liquidity risk and

Liquidity risk other limits of
changes arbitrage imply that
average the price fluctuates
price level around its
conditional mean

Market price
Backtest
• Backtest: Simulation of a trading strategy

• Components of a backtest:

1. Universe
 The universe of securities to be traded
2. Signals
 The data used as input, the source of the data, and how the data are analyzed.
3. Trading rule
 How you trade on your signals, including how frequently you review them and rebalance your
positions, and the sizes of your positions.
4. Time lags
 Making sure the trading rule is implementable
Trading Rules
• Portfolio rebalance rule.
― A portfolio rebalance rule looks at the entire portfolio of securities and defines how it is rebalanced.
― This type of trading rule is backtested as follows. For each time period,
 Determine the optimal portfolio of securities.
 Make a (paper) trade to rebalance to this portfolio.

• Enter–exit trading rule.

― Another approach is to think in terms of discrete trades:
 For each asset, determine when to enter a new trade and how to size the initial position.
 Determine how the position is resized over time, depending on the circumstances.
 Determine when to exit the trade.
Data Mining and Biases
• Backtests typically look a lot better than the real-world trading performance:
― The world is changing
― All backtests suffer from data mining biases
• Avoidable biases:
― biased universe of securities
― biased trading signals and trading rules
― in-sample test (you should do an out-of-sample test )
• Unavoidable biases, e.g.:
― you gravitate toward a version of the trade that has worked well in the past
― you tried the backtest because you heard someone made money on this trade
• Discounting backtests
• Goal:
― A strategy that works well in future trading
― Not simply to have the best possible backtest
Adjusting Backtests for Trading Costs
• Transaction costs reduce the returns of a trading strategy
• Taking transaction costs into account in backtest:
― Compute the return on the portfolio
― Compute the new security positions and the implied trades
― Compute the expected trading costs for every security and add them up
― Subtract the total expected trading cost from the portfolio return
• T-costs are more important:
― the higher the turnover of the trading strategy
― the larger the hedge fund

• Transaction costs change the your optimal trading rule

― Later topic: optimal trading
Backtests and Regressions

Metatheorem. Any regression can be expressed as a portfolio sort, and any portfolio sort can be
expressed as a regression:
• A time series regression corresponds to a market timing strategy
• A cross-sectional regression corresponds to a security selection strategy
• A univariate regression corresponds to sorting securities by one signal, while a multivariate regression
corresponds to double-sorting securities by two signals, allowing you to determine whether one signal
adds value beyond the other.
Market Timing and Time Series Regressions
• Time series regression of the excess return Re of one security, say, the overall stock market, on a
forecasting variable F, say, the dividend-to-price ratio:

𝑒𝑒
𝑅𝑅𝑡𝑡+1 = 𝑎𝑎 + 𝑏𝑏𝐹𝐹𝑡𝑡 + 𝜀𝜀𝑡𝑡+1
• The estimate of the regression coefficient b is given by
∑ 𝐹𝐹𝑡𝑡 −𝐹𝐹� 𝑅𝑅𝑡𝑡+1
𝑏𝑏� = 𝑡𝑡 = ∑𝑇𝑇𝑡𝑡=1 𝑥𝑥𝑡𝑡 𝑅𝑅𝑡𝑡+1
∑𝑡𝑡 𝐹𝐹𝑡𝑡 −𝐹𝐹� 2
• which is the cumulative return on a long/short timing strategy, where the trading position x is given by
𝑥𝑥𝑡𝑡 = 𝑘𝑘 𝐹𝐹𝑡𝑡 − 𝐹𝐹�
• where
𝑘𝑘 = 1/ ∑𝑡𝑡 𝐹𝐹𝑡𝑡 − 𝐹𝐹� 2
Security Selection Strategies and Cross-Sectional Regressions
• Cross-sectional regression with a forecasting variable 𝐹𝐹𝑡𝑡𝑖𝑖 for every security i:

𝑖𝑖
𝑅𝑅𝑡𝑡+1 = 𝑎𝑎 + 𝑏𝑏𝐹𝐹𝑡𝑡𝑖𝑖 + 𝜀𝜀𝑡𝑡+1
𝑖𝑖

• We can run this regression across securities at any time t:

𝑇𝑇
∑𝑖𝑖 𝐹𝐹𝑡𝑡𝑖𝑖 − 𝐹𝐹�𝑡𝑡 𝑅𝑅𝑡𝑡+1
𝑖𝑖
𝑏𝑏�𝑡𝑡 = 2 = � 𝑥𝑥𝑡𝑡𝑖𝑖 𝑅𝑅𝑡𝑡+1
𝑖𝑖
∑𝑖𝑖 𝐹𝐹𝑡𝑡𝑖𝑖 − 𝐹𝐹�𝑡𝑡 𝑡𝑡=1
• Long/short security selection strategy with position in security i is

𝑥𝑥𝑡𝑡𝑖𝑖 = 𝑘𝑘𝑡𝑡 𝐹𝐹𝑡𝑡𝑖𝑖 − 𝐹𝐹�𝑡𝑡

2
• with 𝑘𝑘𝑡𝑡 = 1/ ∑𝑖𝑖 𝐹𝐹𝑡𝑡𝑖𝑖 − 𝐹𝐹�𝑡𝑡 .
Security Selection Strategies and Cross-Sectional Regressions
• The overall estimate of the regression coefficient 𝑏𝑏� using Fama–MacBeth (1973):
𝑇𝑇
1
𝑏𝑏 = � 𝑏𝑏�𝑡𝑡
�
𝑇𝑇
𝑡𝑡=1
• This is the average profit of the strategy over time.
• The risk of the strategy: the volatility of the regression coefficients:
𝑇𝑇
1 2
𝜎𝜎� = � 𝑏𝑏�𝑡𝑡 − 𝑏𝑏�
𝑇𝑇 − 1
𝑡𝑡=1

• Sharpe ratio of the security selection strategy is

𝑏𝑏�
𝑆𝑆𝑆𝑆 = �
𝜎𝜎
• Corresponds closely to the t-statistic of the regression estimate:
𝑏𝑏�
t−statistic = 𝑇𝑇
𝜎𝜎�
Several Trading Signals

• We can regress returns on several trading signals, say, F and G:

𝑖𝑖
𝑅𝑅𝑡𝑡+1 = 𝑎𝑎 + 𝑏𝑏 𝐹𝐹 𝐹𝐹𝑡𝑡𝑖𝑖 + 𝑏𝑏 𝐺𝐺 𝐺𝐺𝑡𝑡𝑖𝑖 + 𝜀𝜀𝑡𝑡+1
𝑖𝑖

• What does bF mean here?

• Similar to double-sorting securities by both F and G each period
• One advantage of regressions is that it is easy to add many variables on the
right-hand side, while is becomes impracticable to quadruple-sort securities.
Multiple Predictors: Example 1
• Suppose that high B/M firms have mean return 12%, while low B/M have 7%
• Suppose that high MOM firms have mean return 11%, while low MOM have 8%
• Should you do both trades?
• If so, will doing both trades be riskier or safer than each separately?
Multiple Predictors: Example 1

Low B/M High B/M

Low E(R)=7% E(R)=12% 8%

MOM #firms=400 #firms=100

High E(R)=7% E(R)=12% 11%

MOM #firms=100 #firms=400
7% 12%

48
Multiple Predictors: Example 2
• Suppose that high B/M firms how mean return 13.8%, while low B/M have 8.2%
• Suppose that high MOM firms how mean return 10.6%, while low MOM have 11.4%
• Should you necessarily short high MOM?
Multiple Predictors: Example 2

Low B/M High B/M

Low E(R)=5% E(R)=13% 11.4%

MOM #firms=100 #firms=400

High E(R)=9% E(R)=17% 10.6%

MOM #firms=400 #firms=100
8.2% 13.8%
Multiple Predictors: Example 3

Low B/M High B/M

Low E(R)=3% E(R)=13% 8%

MOM #firms=250 #firms=250

High E(R)=9% E(R)=17% 13%

MOM #firms=250 #firms=250
6% 15%
Fundamentals

This class

52
Portfolio Construction
Portfolio construction: choosing and sizing the various trades to achieve a good trade-off between risk
and expected return
1. Diversification
2. Position limits and risk limits:
― At the level of securities, asset classes, and overall portfolio
3. Larger bets on higher conviction trades
4. Size bets in terms of risk
5. Correlations matter
― E.g., for a long position, correlation with other longs is bad, corr. with shorts is good
― E.g., powerful to go long/short within each industry, diversify across industries
6. Resize positions according to forward-looking risk and conviction
― I am in this trade with both feet and there is no turning around now – student
― “Hold” is not on option – Lee Ainslie
― A trader must have no memory and forget nothing – trader saying

53
Portfolio Optimization
• Mean–variance approach:
― Portfolio: x = (x1, … , xS)
― Excess returns: Re = (Re,1, … , Re,S)
― γ: risk aversion coefficient

• Objective:
― Future wealth future wealth = 𝑊𝑊 1 + 𝑅𝑅 𝑓𝑓 + 𝑥𝑥 1 𝑅𝑅𝑒𝑒,1 + . . . +𝑥𝑥 𝑆𝑆 𝑅𝑅𝑒𝑒,𝑆𝑆
― Maximize future wealth while limiting risk:
𝛾𝛾
max 𝐸𝐸 𝑥𝑥 ′ 𝑅𝑅𝑒𝑒 − 𝑣𝑣𝑣𝑣𝑣𝑣 𝑥𝑥 ′ 𝑅𝑅𝑒𝑒
𝑥𝑥 2
• Using the variance-covariance matrix Ω, this can be written as
𝛾𝛾
max 𝑥𝑥 ′ 𝐸𝐸 𝑅𝑅 𝑒𝑒 − 𝑥𝑥 ′ Ω𝑥𝑥
𝑥𝑥 2

• By differentiating, we get the first order condition:

0 = 𝐸𝐸 𝑅𝑅 𝑒𝑒 − 𝛾𝛾Ω𝑥𝑥
• Optimal portfolio:
𝑥𝑥 = 𝛾𝛾 −1Ω−1𝐸𝐸 𝑅𝑅 𝑒𝑒

54
Benefits of Portfolio Optimization
• This optimal portfolio
― is characterized by taking large positions for securities with large expected returns, low variance,
and low correlation to other long positions
― Portfolio optimization helps reduce people’s behavioral biases,
 Tendencies to make certain mistakes
 For instance, people like to hang on to their losing positions even if the reason they liked the
securities no longer applies, and they like to sell winners to lock in gains even if the trade has
gotten even better.
• Limitations. The “optimal” portfolio is often problematic in practice:
― Risk and expected returns are estimated with errors, and often using different techniques
― Real-world portfolios are often subject to a number of constraints on position sizing and trade
sizes, and, while these constraints can be added to the problem, they often distort the solution
― Need to account for transaction costs
― Hedge fund trades repeatedly

55
Measures of Risk
• Volatility
• Value at Risk (VaR)
― The maximum loss with a certain confidence
― E.g. 95% confidence, or 99% Probability density

5% probability

ES Return
Value at Risk
• Expected shortfall: ES = E( loss | loss > VaR )

56
Measures of Risk
• Stress loss and stress tests
― Simulated portfolio returns during various scenarios
― E.g. significant past events
 1998 price shocks around the LTCM bailout
 September 11,
 The failure of Lehman
― Imagined future events:
 The failure of a sovereign (e.g., Greece)
 A large interest rate move
 A large shock to equity prices, a spike in volatility
 A sharp increase in margin requirements
• Stress tests explore cases where you do not have enough data to estimate the risk accurately, as well
as events that can play out over several days
― what actually happens during a crisis never corresponds exactly to any stress tests, but one hopes
that preparing for foreseeable events will provide the discipline to survive what actually happens

57
Managing Risk I: Prospective Risk Control
• Sound portfolio construction is first line of defense
• Strategic risk target
― Average level of risk that the fund intends to take over the long term
― For instance, the strategic risk target could be measured as the fund volatility
― Often ranges from bond-like volatility to equity-like volatility, say, somewhere between 5% and 25%
annualized volatility
• Tactical risk target
― Tactical risk varies around the strategic risk target, depending on the
 Investment opportunities
 Market conditions
 Recent performance

58
Managing Risk II: Reactive Risk Control Drawdown Control
• A hedge fund may want to minimize the risk that its drawdown will become worse than its maximum
acceptable drawdown (MADD), say, 25%.
• If the current drawdown is given by DDt, then one sensible drawdown control policy is:
VaRt ≤ MADD – DDt
• Or, alternatively, for some number s and volatility σt,
s × σt ≤ MADD – DDt
• If this inequality is violated, the hedge fund should reduce risk
• Later, as strategy recovers from losses, risk is increased

59
Benefits of Drawdown Control
• Benefits of have a planned drawdown control
― Reduce risk earlier
― Increase risk earlier
• Reducing risk after losing on a position is painful
― The trader feels that a loss is being locked in if he or she unwinds
― Many related sayings:
 Your first loss is your least loss
 As a trader, never panic, but if you are going to panic, panic first
 The strongest weak hand suffers the largest loss
• Why do people tend to sell at
the bottom in a crisis?

60