The Q-Group

Spring 2015 Seminar

The Ritz-Carlton

Amelia Island, Florida

March 30-April 1, 2015

Reviews by David Modest and Russ Wermers

**Executive Summary**

Florida finally warmed to Q Members through great weather during the latter days of the Spring Seminar. Located next to a stunning Atlantic Ocean white sand beach, with an abundance of seashells evident to anyone who walked or jogged near the water, several academic speakers discussed different aspects of the relation between risk and reward-mainly focusing on equity markets.

“Get active, but don’t move too much”! **Martijn Cremers** opened the Q meeting with a surprising

finding: funds with high Active Share only outperform when they don’t trade very much. The idea is to look for active managers who take long-term bets away from a passive index.

**Scott Murray** followed by reexamining an age-old question: why is the Security Market Line so flat (i.e.,high beta stocks do not perform as expected)? He argued that the real underlying anomaly is that high beta stocks tend to be those with extreme daily returns, which make them attractive to individual investors who are attracted to “lottery-like” payoffs (i.e., gambling behavior).

**Robert Whitelaw** followed by discussing what can be done by an equity manager to hedge against

interest rate increases. Surprisingly, the answer is to invest in stocks with high leverage!

The next talk took on an international flavor. First, **Adrien Verdelhan** presented a new version of the International CAPM. Verdelhan showed that a three-factor model, the “Redux” model, can explain a significant amount of variation in the returns of a wide variety of passive developed and emerging market equity indices, as well as the returns of a sample of actively managed mutual and hedge funds.

Q Members enjoyed a break from the academic papers when they attended dinner: Our own **Jack Treynor** was honored through remarks delivered by **Joanne Hill** and Marc Reinganum. Joanne and Marc cited a few of Jack’s astonishing career accomplishments, and then Jack rose to deliver a short address. While his complete remarks can be found in this document, the main takeaway was that Jack challenged social scientists, including financial economists, to borrow some of the risk-taking creativity of the natural sciences, where breathtaking discoveries seem to occur almost every month!

The morning talk on Day 2 also took an international flavor, as **Alberto Cavallo** discussed the long-lived question of whether purchasing power parity has any predictive content on exchange rates. Cavallo showed how MIT’s Billion Prices Project has collected impressive data on the pricing of the same good in many different countries. With these new data, he showed some persuasive evidence that relative PPP (prices may differ due to frictions, but the gap between countries stays the same) appears to hold.

**Ian Cooper** followed with empirical results that indicate that investor sentiment does not appear to affect deviations from fundamental value among upstream oil stocks. Cooper used this set of stocks because they present a better opportunity to be valued from fundamentals, that is, oil prices. Instead, Cooper found that any effect of sentiment on these stock returns came through its effect on the fundamentals themselves!

**Robert Stambaugh** “turned-over turnover” by showing that equity mutual funds with higher turnover

relative to their past subsequently outperform. The key to his analysis is that he controls for the tendency of a particular fund to trade a lot (or not), and examines the results when it changes its trading activity over time.

Next, **Chester Spatt** reexamined the 15-year-old puzzle of why 3COM shares seemingly did not reflect

the value of their ownership of Palm shares after the spin-off of 5% of Palm from 3COM. Spatt demonstrated that, once one properly accounts for the value of holding Palm shares, which includes a large expected yield from lending them to arbitrageurs, then the mispricing of 3COM disappears.

“Who’s On First?” seemed to be the question that **Anna Scherbina** opened with on Day 3. She used

Granger Causality regressions to determine which stock, following a news event, is the leader stock, and which stocks have returns that “follow the leader”. Scherbina surprisingly found that “leader stocks” can both be smaller in size and can be from different industries from their “followers”!

**Kent Smetters** wrapped up the sessions with a talk that sharpened the Sharpe Ratio. Smetters derived this new “Sharper Ratio” starting with a general utility function for an investor, and then showed how the

optimal measure of performance can be derived for an investor with a certain level of risk-aversion.

After this set of inspiring and thought-provoking papers, Q members left the Spring meetings with one last question firmly in mind: “Will Zona be as nice for our meeting this October?”

Martijn Cremers, Professor of Finance at the University of Notre Dame, built on his prior well-known paper with Antti Petajisto that introduced “Active Share”. Active Share (AS) uses the portfolio holdings of a fund to compute the average absolute distance of weights from the “best fit benchmark” for that fund. In that paper, AS was shown to predict fund performance over the next several months.

In his new paper with Pareek, Cremers more deeply investigates the role of AS in predicting fund performance by separating funds in a second dimension: the holding duration of stocks. For instance, one can imagine two extreme types of high AS managers. The first trades a lot, which results in holding portfolios that vary a lot over time, but that usually deviate from the benchmark. The second type are “patient managers,” who trade very little, holding stocks potentially for years at a time, but also in weights that deviate significantly from the benchmark.

The research question that Cremers and Pareek (CP) address is “which of these two types of high AS managers performs better?” This is an important follow-on question to the AS paper, as it is natural to think of a high AS manager as having high portfolio turnover-but, this need not be the case! In addition,Cremers stated that over the past 10 years (since 2001), the highest AS funds have not outperformed their benchmarks; therefore, the AS measure is insufficient, by itself, to find outperforming fund managers.

Thus, CP suggest adding their measure of the holding period of a stock, the “duration measure,” as a second conditioning variable” in addition to the AS of a fund manager. What is the duration measure? It is simple: it is the average length of time a dollar in a manager’s portfolio has been invested in the same stock (over the past 5 years). For example, in 2010, Berkshire Hathaway had a duration measure of roughly 4 years, while Fidelity had a measure of about 2 years.

Cremers conducts his tests of AS combined with duration with 3 samples: all-equity retail mutual funds

(net returns), all institutional portfolios (long-only 13(f) data), and hedge funds (long-only 13(f) data). As

a side note, Cremers recommends a new easier-to-compute version of AS:

where the summation in the above expression is across all overlapping positions.

Cremers showed that U.S. equity mutual funds ranked in the top quintile by their AS measures have very different following-year alphas (using a four-factor Fama-French plus momentum model, plus a liquidity factor), depending on each fund’s duration measure. High AS funds with a long duration measure (those that deviate strongly from the benchmark, but have low turnover) produced an alpha that exceeded 2%/year, while short duration funds produced an alpha below -2%/year (see the red bars below):

1 Further, recent white papers by Fidelity (Cohen, Leite, Nielson, and Browder, 2015) and AQR (Frazzini, Friedman,

and Pomorski, 2015) have criticized the efficacy of Active Share.

What explains the outperformance of these high AS, long duration funds? Cremers finds that about half of the alpha can be explained through the tendency of such funds to invest in high quality stocks (those that are profitable and growing).

Finally, Cremers showed that high AS, long duration funds also outperform when we examine all 13(f) institutional investors, as well as when we examine hedge funds (that file 13(f) statements).The takeaway for Q members is that Active Share, by itself, is not sufficient to locate skilled managers. One must also find patient managers!

__Bibliography__

Cohen, Tim, Brian Leite, Darby Nielson, and Andy Browder, 2015, Active Share: A Misunderstood Measure in Manager Selection, Investment Insights (Fidelity Investments), available at https://www.fidelity.com/bin-public/060_www_fidelity_com/documents/leadership-series_activeshare

Cremers, Martijn, and Ankur Pareek, 2014, Patient Capital Outperformance, Working Paper, available at https://www.q-group.org/wp-content/uploads/2014/11/cremers-paper-SSRN-id2498743.pdf .

Cremers, Martijn, and Antti Petajisto, 2009, How active is your fund manager? A new measure that predicts performance, Review of Financial Studies 22, 3329-3365.

Frazzini, Andrea, Jacques Friedman, and Lukasz Pomorski, 2015, Deactivating Active Share, AQR White Paper, available at https://www.aqr.com/library/aqr-publications/deactivating-active-share .

Brown, and Yi Tang

Scott Murray, Assistant Professor of Finance at the University of Nebraska, talked about one of the most enduring and documented anomalies in financial economics: that the empirical security market line (SML) is too flat. That is, high beta stocks generate returns that are lower than that predicted by the SML, while low beta stocks generate returns that are higher. This anomaly was documented as far back as Black, Jensen and Scholes (1972), and has puzzled researchers for more than four decades. Recently, Frazzini and Petersen (2014) have presented a model and results arguing that leverage constraints cause lower risk assets to outperform higher risk assets (on a risk-adjusted basis) across many asset classes,including equities, sovereign and corporate debt, and futures. The idea is that leveraged-constrained investors who wish to hold a portfolio that tracks the market (i.e., beta=1) must hold stocks centered around that level-reducing the demand for low-beta stocks.

In his talk, Scott Murray presented evidence of an alternative driving force for the betting against beta effect (defined as investors outperforming by holding low-beta stocks). Murray argued that, in U.S. equity markets, the demand for high-beta can be explained by the demand for stocks with lottery-like payoffs (lottery demand), proxied by the average of the five highest daily returns of a stock in a given month (henceforth, MAX). In short, high beta stocks also tend to have a high value of this lottery demand proxy.

Murray also showed that this phenomenon only exists among stocks with a low proportion of institutional ownership, consistent with lottery demand being concentrated among individual investors. Further, he showed that the lottery demand phenomenon cannot be explained by a variety of firm characteristics (market capitalization, book-to-market, momentum, illiquidity, and idiosyncratic volatility), risk measures (co-skewness, total skewness, downside beta, and tail beta) and funding liquidity measures (TED spread sensitivity, sensitivity to TED spread volatility, T-bill rate sensitivity, and financial sector leverage sensitivity).

The key table (Table 5 in the paper) is shown below. It presents evidence showing that betting against beta phenomenon is subsumed by the lottery demand effect (MAX10 consists of stocks with the highest average of their five daily highest returns during a month). That is, holding beta fixed, the lottery demand effect persists; whereas, the betting against beta phenomenon disappears after holding the lottery demand effect fixed.

The table presents the results of bivariate independent decile sorts based on both beta and the lottery demand proxy. The intersections of each of the decile groups are then used to form 100 portfolios. (As the sorts are independent, the number of stocks in each of the 100 portfolios can differ.) The last two rows of the table show that the beta effect disappears after controlling for the lottery demand proxy — both in terms of the raw return difference between the high and low beta portfolios (High-Low) and the Fama French 4-factor alpha for the difference between the high and low beta portfolios (FFC4); no t-statistic is larger than 1.61 in absolute value. Conversely, the right-most two columns of the table show that the negative relation between the lottery demand proxy and future stock returns persists after controlling for the effect of beta — with t-statistics for the FFC4 alphas ranging from -2.70 to -7.43.

Murray provided insight into the channel by which lottery demand generates the betting against beta phenomenon. He showed that in a typical month, market beta and the lottery demand proxy have a high cross-sectional correlation. In months when this cross-sectional correlation is high, the betting against beta phenomenon is strong, as can be seen in the first three sets of rows in the table below (High). The first row shows the average beta (the sorting variable), while the second and third show the following month return and four-factor alpha, respectively. However in months when the correlation is low, the betting against beta phenomenon disappears. This can be seen in the bottom three sets of rows in the table (Low). Murray’s conclusion is that the betting against beta strategy is a proxy for a much better strategy, which might be coined the betting against lottery demand!

The below graph clearly summarizes the overall idea of Murray’s paper: the empirical SML is flatter only during months when the lottery demand is strong:

Murray closes with one important point: A key difference between the betting against beta and lottery demand explanations of this empirical anomaly has to do with the role of institutional investors in generating the price pressure. Frazzini and Petersen (2014) have argued that it is the leverage constraint of institutional holders that generates the phenomenon, and, hence, the effect should be strongest for those stocks with a high degree of institutional ownership. Murray argued that if the lottery demand explanation is correct, the effect should be strongest for those stocks with low institutional ownership. The following table shows the returns for bivariate dependent portfolios, where stocks are initially sorted into ten deciles based on degree of institutional ownership, then sorted within each decile by beta. As is apparent, the betting against beta phenomenon is actually strongest for those stocks with the lowest institutional ownership (column 1: INST 1) and weakest for those stocks with the highest institutional ownership (column 10: INST 10) — additional evidence in favor of the lottery demand explanation.

__Bibliography__

Bali, Turan G., Stephen Brown, Scott Murray, and Yi Tang, Betting against Beta or Demand for Lottery,

2014, available at https://www.q-group.org/wp-content/uploads/2014/11/BaliBrownMurrayTangPaper.pdf

Black, Fischer, Michael Jensen, and Myron Scholes, 1972, The Capital Asset Pricing Model: Some Empirical Tests. Available at http://ssrn.com/abstract=908569.

Choi, Matthew Richardson, and Robert Whitelaw

Robert Whitelaw, Edward C. Johnson III Professor of Entrepreneurial Finance at NYU, and CIO of Index IQ, presented a paper on the relation of stock returns to movements in interest rates. This topic is of great interest, as the risk of interest rate increases brings uncertainty to investors about how their stock portfolios will be affected.

The surprising take-away from Whitelaw’s lecture was that some stocks can actually serve as an effective hedge against interest rate increases, and these stocks can be identified by their high level of debt in the capital structure (high leverage)!

To understand this, there are two effects of the impact of changes in interest rates to a company’s balance

sheet to understand:

1. Change in interest rates » change in asset value » change in the value of debt and change in the

value of equity (total pie gets bigger or smaller)

2. Change in interest rates » change in the value of debt relative to the change in the value of

equity (the changes are different, thus, the share of the pie taken by debt vs. equity changes)

Since the impact on the value of debt is negative from an increase in interest rates, the change in the value of equity can be positive or negative, depending on the change in asset value as well as the leverage of the firm. All else equal, higher leverage leads to a more negative duration of equity-meaning that equity can benefit from interest rate increases! This can also explain (partially) the time-varying correlation between stock and bonds (sometime they move together, and other times they don’t).

Whitelaw next takes this theory to the data. He maps out each firm’s capital structure, and then constructs returns for each security (no small feat for the bonds!). The return on the asset side of the balance sheet is

computed as the value-weighted average of the returns on the liability side (equity plus debt). Whitelaw computes duration estimates (presented below) on each tranche of the capital structure for firms with different levels of leverage:

Note two effects here. First, within each leverage “bucket” (each row), duration increases with priority of the claim in the capital structure. Thus, senior debt is most adversely impacted by a rise in interest rates, followed by junior debt, and then equity. Second, firms appear to choose higher leverage when the duration of their assets is higher (i.e., when the value of the assets is more affected by changes in rates), as can be seen in the far right column above. This means that firms attempt to hedge some of the interest rate risk that could potentially be borne by their equity holders-adding more debt, which takes the brunt of the increased interest rate risk of the higher duration assets!

Whitelaw further shows that, once you control for the change in asset duration with increases in leverage,the duration of equity becomes more negative as leverage increases.

Whitelaw concludes with another insight. Bond portfolio returns are often modeled as a function of “term” and “default” factors, where the default factor is proxied by the difference in returns between a long-term corporate bond and a long-term government bond, such that the maturities are effectively matched. Whitelaw argues that this construct of the default factor is misspecified. That is, it does not properly isolate default risk, because the corporate bond duration can be very short if they are low priority (or highly leveraged) bonds in a firm’s capital structure, and, hence, the bonds underlying the default factor, while maturity matched, are not duration matched.

The takeaway for Q members: use a default factor that is corporate bond returns minus short-term government bonds to better isolate default risk!

__Bibliography__

Choi, Jaewon, Matthew Richardson, and Robert Whitelaw, 2014, On the Fundamental Relation between

Equity Returns and Interest Rates, available at https://www.q-group.org/wp-content/uploads/2014/11/Whitelaw-paper.pdf.

Adrien Verdelhan, Associate Professor of Finance at MIT, revisited the question of whether foreign currency risk is priced in global equity markets. He began his talk by reviewing the two primary international models of equity pricing: the World CAPM and the International CAPM. Under the assumptions of the World CAPM, purchasing power parity (PPP) holds instantaneously, and global equity market risk is the single source of systematic risk. Sensitivity (beta) to that risk determines all equity expected returns, and currency risk is irrelevant from a pricing perspective. In contrast, the International CAPM presumes that PPP does not hold instantaneously. As a consequence, all bilateral exchange rates are additional systematic risk factors that theoretically impact all expected asset returns.

The ICAPM holds under very general assumptions. However, it is difficult to implement practically, as it requires the use of all bilateral exchange rates. Verdelhan presented a parsimonious reduced form model (the CAPM Redux) that assumes that: (i) the existence of a pricing kernel that insures that the law of one price holds across assets, and (ii) investors can form portfolios freely-going both long and short.

Given these assumptions, the expected return on any asset can be written as a time-varying function of three factors: the average world equity market return denominated in local currency terms, a carry factor that is the excess return of a portfolio that goes long high-interest currencies and short low-interest currencies, and a dollar factor that presumes, each period, that an investor borrows in the U.S. and invests equally in all other currencies.

Verdelhan runs a variety of horse races to compare the performance of the CAPM Redux to its three main competitors: the World CAPM, the International CAPM, and an international version of the Fama French four factor model. The most persuasive evidence is provided in four graphs (one for each model) of realized average excess returns, plotted on the y-axis, against the corresponding model predicted returns, plotted on the x-axis. The points in each graph correspond to over 200 different indices in 46 countries, where, for each country, returns are computed for five different MSCI equity indices: the aggregate stock market, growth, value, large market-capitalization, and small market-capitalization. As is readily apparent, both the World CAPM and the International CAPM generate virtually no cross-sectional variability in expected returns. While a good model would be characterized by points randomly distributed around the 45 degree line (shown in each graph), most of the data points for these two modelsare best described by a vertical line.

In contrast, both the Fama-French four factor model and the CAPM Redux do a much better job generating cross-sectional variability in expected returns. The data points for the CAPM Redux are the most tightly clustered around the 45 degree line, indicating that it is the best model

Verdelhan then provided evidence on a wide variety of model robustness, performance, and relevance tests for the CAPM Redux model. He showed the importance of time-varying betas for all three factors in that model, the importance of the carry and dollar factors in explaining mutual fund and hedge fund returns, and the superior performance of the CAPM Redux model with respect to rolling mean absolute alphas-especially in the last decade (depicted in the graph below). Specifically, the below graph shows alpha estimates (in absolute value) averaged across passive country equity indexes at each point in time. Verdelhan argued that that his Redux model is superior because it delivers lower mean absolute alphas (in other words, it explains passive country index returns best, and does not assign high alphas to them).

In short, he was able to show that a parsimonious three-factor model can explain a significant amount of

variation in the returns of a wide variety of passive developed and emerging market equity indices, and in

the returns of a sample of actively managed mutual and hedge funds.

__Bibliography__

Brusa, Francesca, Tarun Ramadorai, and Adrien Verdelhan, 2014, The International CAPM Redux available at

https://www.q-group.org/wp-content/uploads/2014/11/verdelhan-paper-SSRN-id2520475.pdf

**Monday Evening Dinner**

Rather than the usual evening keynote, Q members were treated to a special event-the recognition of one of our most valued members – Jack Treynor!

Jack has been a valued wise elder of Q for quite some time, and Q members can easily spot him by the cap he wears. Jack has been a member of the Q Research Committee since 1975, an astounding 40 years! Jack has also served on the Q-Group Board for decades.

Joanne Hill and Marc Reinganum opened the dinner presentation with a tribute to Jack. In their remarks, they discussed just a few of Jack’s numerous contributions to the field of finance. For Q members, Jack’s research has clearly and profoundly influenced our chosen profession, quantitative asset management! Indeed, Joanne said that Jack is one of the founders of quantitative investment management, and that he helped to align investment management practice with the best of academic research.

And, of course, Jack worked on a capital asset pricing model in the early 1960s, creating the earliest drafts of ideas of what would eventually become the CAPM. He also worked on several papers with Fischer Black, including the famous Treynor and Black (1973) paper on the information ratio that showed how to optimally construct a portfolio of risky assets, and that still wields incredible influence on the investment management world today.

Q members should note that, in his many years of participation, Jack presented papers at the Q-Group on 26 occasions-a “World Record”!

Jack Treynor then rose to talk to Q members. In a poignant and brief set of remarks, Jack recognized the influence of Operations Research-his field-on Modern Portfolio Theory. He noted that Bill Sharpe and Harry Markowitz started their careers in Operations Research. He also pointed out that Operations Research has brought the values and attitudes of the physical sciences to the social sciences. Jack closed with a clear mission for social scientists: In the physical sciences, exciting new discoveries are made each month. Jack challenged us to make exciting new discoveries in finance each month, as well!

**Here are Jack’s remarks:**

- Are Markowitz and Sharpe so productive because of their years in operations research?
- Every month there are new discoveries in the physical sciences. Every month there will be a life-saving discovery in molecular biology that will change the practice of medicine.Next month, there will be another exciting discovery about earth-like planets. But, next month, there will be no exciting discoveries in the social sciences.
- When I was young, I thought operations research was just about applying the techniques of the physical sciences to social problems. Now that I am deep into retirement, I think operations research is about applying the research philosophy of the physical sciences to social problems.
- But why would research philosophy matter? The philosophy of the physical sciences changed at the beginning of the 17th century. The discoveries that followed made the Industrial Revolution possible.
- But what was the difference between the old research philosophy and the new? The old philosophy valued erudition; the new philosophy valued discovery. But why were their objectives mutually exclusive?
- Because one is the Type 1 error and the other is the Type 2 error, and they are mutually exclusive: if you want to avoid Type 1 errors, commit Type 2 errors; if you want to avoid Type 2 errors, commit Type 1 errors.
- Type 1 errors reject as false ideas that are true; Type 2 errors accept as true ideas that are false. Because it valued erudition, the old philosophy minimized Type 1 errors; because it valued discovery, the new philosophy minimized Type 2 errors.
- The physical sciences have never forgotten the lesson of the Industrial Revolution; the social sciences are too young to remember it. And that’s why they are mutually exclusive research philosophies.
- Aren’t these two different cultures?
- Don’t the cultures require their members to have the right research philosophy?
- If you have grown up in one culture, isn’t it consequently hard to change?
- If academics value erudition, are they less likely to challenge the old ideas?
- Are people whose work brings them face-to-face with real-world problems going to be aware of the need for change?
- Are they likely to be less impressed with erudition?
- Are central bankers the kind of practitioners who would be stimulating for academice-conomists?
- Do they prefer problem solving to erudition?
- Wouldn’t a problem-solving culture be useful for professional investors?

__Bibliography__

Treynor, Jack L. and Fischer Black (1973). “How to use Security Analysis to Improve Portfolio Selection”. Journal of Business 46, No.1, pp. 66–86.

Alberto Cavallo, Associate Professor of Applied Economics at MIT Sloan, presented research based on a unique set of data collected daily from hundreds of on-line global retailers. The initial data were obtained as part of MIT’s Billion Prices Project that was started in 2008 with Roberto Rigobon, an MIT colleague.

The data collection makes use of the HTML code that underlies public webpages, and it “scrapes” this information to extract price and product information for all the products the retailers sell. The data set has many attractive features relative to other sources of purchasing power parity (PPP) data: It primarily focuses on tradable goods, is available for many countries, is comparable across time, is available for hundreds (and eventually thousands) of products, can track dozens of varieties of the same product, provides exact product matches, and is available on a daily basis. The data also has the advantage that PPP has precise and refutable predictions with regards to price levels, whereas price index data (such asthe CPI) can only be informative about price changes.

Cavallo’s presentation focused on several different aspects of this research project. First, he reported on the research that they conduct to verify the reliability of the online price data. This involved the simultaneous random sampling of online and in-store prices using a bar scanning app and crowd-sourced workers. Table 2 below shows high degrees of similarity between the prices scraped from online websites and the actual prices charged in stores for most countries, with the notable exception of Argentina.

The main focus of the talk was on the behavior of the real exchange rate (RER) defined as:

where is Plc the price of a good in the local currency, Pus is the price in US dollars, and Eusd/lc is the exchange rate in US dollars per unit of local currency. If absolute PPP always holds then RER=1 at all

times. If, instead, relative PPP always holds, then RER=fixed constant.

The empirical literature has found that absolute PPP does not hold, and that relative PPP only holds in the long run-with RER shocks having half-lives in the 3-5 year range. Cavallo reported on research that revisits the empirical validity of absolute and relative PPP using more accurate micro-level price data. Cavallo noted two primary research objectives: 1) measuring the level of the RER over time, and 2) estimating how fast the RER returns to steady state after a shock.

Cavallo first presented results using Brazilian data. As can be seen in the figures below, the RER in Brazil, relative to the U.S. (light blue line), is much more stable than relative prices (red line, left axis) or the nominal exchange rate (dark blue line, right axis). This negative correlation between relative prices and nominal exchange rates was apparent in 6 of the 7 countries for which data was presented; the sole exception was China, where the currency does not freely float.

He then focused on Argentina. Over the last four years, the RER in Argentina has typically been in the 1.00 to 1.10 range. However, over the last year, the RER has risen from 1.00 to roughly 1.20-suggesting that, over time, the currency will have a significant depreciation.

Cavallo presented summary evidence for seven countries based on a vector error correction model (reproduced below), where the coefficient B0=0 if absolute PPP holds, B1=-1 if relative PPP holds, and the ratio indicates how much of the adjustment comes through relative price changes compared

to changes in nominal exchange rates. A ratio less (greater) than one suggests most of the adjustment comes through changes in nominal exchange rates (relative prices). The table also contains estimates of the half-lives (measured in days) of the relative price and exchange rate changes. In contrast to previous research, Cavallo finds that real exchange rate adjustments are much faster than previously estimated: largely occurring within months rather than years, as previously thought using CPI data. The adjustment mechanism and speed vary by country. Thus, the key to Cavallo’s new findings that prices and exchange rates adjust faster than previously thought is the precise cross-country pricing data for the same goods that he collects from online postings.

__Bibliography__

Cavallo, Alberto, PPP and Exchange Rates: Evidence from Online Data in Seven Countries, MIT Working Paper.

Whose “sentiment” affects stock prices-the sentiment of professional investors, or that of retail investors? Ian Cooper, Professor of Finance, London Business School, tackles this question though an analysis of the stocks of upstream oil companies.

In order to study sentiment, one must first clearly understand the fundamental value of a stock. Cooper argued that the relation of upstream oil company stocks to fundamentals are clearly defined, in theory, because of their simple tie to oil prices. That is, if the Hotelling (1931) Rule is correct, the value of such companies equals the current value of their reserves minus extraction costs (another way to express this is that the price of an exhaustible resource, minus extraction costs, should rise over time at the risk-free.

One can determine the deviation of oil and gas stocks from fundamental value by measuring the deviation of equity prices from reserves minus extraction costs (minus the value of debt). Specifically, Cooper assumes that fundamental value is a function of the month-end spot price of West Texas Intermediate oil and the spot wellhead price of West Texas natural gas (as well as the contango in oil prices).

Cooper uses these 121 U.S. Oil and Gas E&P stocks during 1983-2011 to test the effect of investor sentiment on stock returns. That is, he wished to explore whether investor sentiment pushes these upstream oil stock prices away from their fundamental values in the short-term (Barberis, et al., 2012) or whether sentiment predicts future returns because of their impact on fundamentals themselves.

Cooper measures sentiment in two different ways: the Baker and Wurgler (2006) “Sentiment Index,” and the proportion of individual investors who report that they are bullish in the regular survey conducted by the American Association of Individual Investors (“Bullishness” survey). The monthly correlation between these two sentiment metrics is only 9%, indicating that they capture different aspects of sentiment.rate).

Accordingly, Cooper’s paper assumes that stock prices are affected by the actions of two types of traders:

a professional arbitrageur, who sells when the Baker-Wurgler sentiment measure is high, and a naÃ¯ve trend-follower, who buys when the AAII retail sentiment measure is bullish. Then, he allows, in separate regression equations, fundamentals and non-fundamentals, to depend on the actions of both types of traders (through the level of the two sentiment variables). That is, Cooper regresses, separately,fundamental returns and non-fundamental returns on the two lagged sentiment measures. These regression equations are combined in a vector autoregressive model, so that fundamental returns can alsodepend on lagged non-fundamental returns (and vice-versa).

Next, to obtain his y-variables in the above system, Cooper splits upstream oil stock returns into fundamental and non-fundamental returns by first regressing monthly returns over a 60 month period on the change in WTI prices, natural gas prices, and the contango (logged 6th minus 1st futures price) of WTI:

The fitted regression is then used to generate the fundamental returns, while the residuals from this regression are the non-fundamental portion of returns.

Cooper’s results show that fundamental returns, but not non-fundamental returns, are predicted by both of the lagged sentiment indicators. Cooper finds that this predictive role of sentiment on fundamentals is 20 only present starting in 2000, which is roughly when investors became interested in the oil sector (and commodities, in general).

Further, Cooper finds that the Baker and Wurgler index predicts mean-reversion in upstream oil stock prices, while the AAII index predicts momentum in stock prices.

The key takeaway from Cooper’s study of upstream oil company stock prices: sentiment seems to work mainly on fundamentals themselves, and does not appear to drive prices away from fundamental values!

__Bibliography__

Baker, Malcolm, and Jeffrey Wurgler, 2006, Investor Sentiment and the Cross-section of Stock Returns, Journal of Finance 61, 1645-1680.

Brealey, Richard, Ian Cooper, and Evi Kaplanis, 2015, The Behaviour of Sentiment-Induced Share

Returns: Measurement when Fundamentals are Observable, available at

https://www.q-group.org/wp-content/uploads/2015/04/oil-paper-Q-GROUP.pdf.

Hotelling, Harold, 1931, The Economics of Exhaustible Resources, Journal of Political Economy 39,137–175.

and Lucian Taylor

Robert Stambaugh, the Miller Anderson & Sherrerd Professor of Finance at the Wharton School, presented new research examining the empirical link between mutual fund trading activity and future performance. Many previous papers have looked at aspects related to the cross-sectional relationship between trading activity and performance, including the Cremers paper (“Paper #1” above) at this conference. Cremers, for instance, finds that funds that are more patient (i.e. trade less than other firms in the cross-section) tend to have superior returns to other funds that trade more intensely.

In contrast, Stambaugh focused on the time series relationship between trading activity and future performance for each fund. In particular, he examined whether periods of above-average trading activity (relative to a time series mean of trading activity for that fund) result in subsequent periods of aboveaverage returns. The test, conducted on a sample of 3,126 mutual funds during the 1979 to 2011 period, is based on the simple idea that skillful managers should trade more when the opportunity set is more promising.

The primary test of this hypothesis is whether the coefficient b > 0 in the following regression:

basic variants of this regression that have different assumptions about the intercept term: (i) the intercept

is fixed over time and the same for all funds (no fund or month fixed effects), (ii) the intercept can vary

by fund but is fixed over time (fund-only fixed effects), (iii) the intercept can vary by month but is the same for each fund every month (time-only fixed effects), and (iv) there are both month and fund fixed effects. The main results of the paper can be summarized in the following table. The bottom row of the table shows that, once one allows fund fixed effects (i.e. the intercept can differ across funds), the coefficient b is estimated to be positive and very statistically significant:

The 0.00123 estimate for b (with a t-statistic of ~6.6) suggests that a one standard deviation increase in turnover is associated with a 0.65% per year increase in performance for the typical fund.

Stambaugh then showed that the relation between turnover and future performance is stronger for smaller funds and funds with high expense ratios. He argued that the results are concordant with a priori theory: the result should be stronger for smaller funds where the alpha potential is less hurt by the decreasing returns to scale in asset management, and the link should be stronger for more talented managers who are likely to command a higher fee.

__Bibliography__

Pastor, Lubos, Robert Stambaugh, and Lucian Taylor, Do Funds Make More When They Trade More? Working Paper, available at https://www.q-group.org/wp-content/uploads/2014/11/stambaugh-paper-turnover5.pdf

Chester Spatt, Pamela R. and Kenneth B. Dunn Professor of Finance, Carnegie Mellon University, and former SEC Chief Economist, discussed the 15-year-old 3Com-Palm mispricing puzzle. The central question addressed was why Palm, a subsidiary of 3Com, traded at a higher price than implied by the stock price of 3Com.

The following time-line provides a summary of the events surrounding the 3Com-Palm puzzle:

The long-standing puzzle, of course, is why the 3Com stock price from March 2 to July 27 did not properly reflect the implied value of its Palm ownership. For example, Lamont and Thaler (2003) wondered whether the market “can add and subtract,” claimed that the above facts meant that the Law of One Price were violated, and showed additional evidence that options were mispriced such that Put-Call Parity did not hold. These phenomena have more widespread implications than simply Palm, as Mitchell, Pulvino, and Stafford (2002) identified 84 cases of negative stub values during 1985-2000.

In his talk to Q members, Spatt addressed two related questions: (1) Did Palm’s share price reflect the fundamental value that could be attributed to all future cash flows associated with owning the stock (including lending fees)? (2) Did the relationship between Palm’s and 3Com’s stock prices violate the Law of One Price? He argued that, if one takes into account the “convenience yield” that ownership of Palm shares provided to its owners through the ability to lend these shares, as well as the uncertainty of when the spin-off would actually happen, then the pricing of Palm and 3Com shares were consistent with the Law-of-One Price and rational markets.

First, some observations. The graph below shows the annualized percentage borrowing rate for Palm shares over time. Note two important things: first, that the borrowing rate was generally extremely high-from 20 to 60% per year. Second, the borrowing rate climbed substantially in early May, right at the time that 3Com resolved exactly when Palm would be spun-off.

Second, the explanation of the negative stub value. Palm shares were in short supply after the 5% spinoff-

with very limited stock loan availability and a small float for potential buyers. The small float created an opportunity for those who controlled Palm shares to charge very high lending rates starting March 2, 2000. To exploit the negative stub value of 3Com, one must borrow shares of Palm to short them, in combination with a long position in 3Com.2 After factoring in the steep borrowing costs, Spatt showed that there was no way to arbitrage the apparent negative stub value.

**A Deeper Dive**

Spatt showed how to quantify the computation of a “rational” stub value for 3Com, and that, properly computed, it was (almost) always positive-presuming there is no wedge between what borrowers would have to pay to borrow the Palm shares and what holders of Palm shares could earn by lending the stock (e.g., there is no large capture of the borrowing cost by brokers).

A key to his approach is noting that Palm shareholders can earn the lending fees on Palm shares, while the

3Com shareholders effectively own a forward contract on the Palm shares, and cannot earn those fees. Thus, the intrinsic value of Palm stock should exceed the present value of a forward contract on the stock by the present value of expected lending fees:

Another salient point is that 3Com may have cancelled the spin-off of the remaining 95% of Palm, and might have

spent the cash flows from Palm on negative NPV projects due to agency problems in corporate management.

We can use this equation to determine the market’s valuation of the STUB. The only problem is that there

is no forward market for Palm shares, but Spatt uses put-call parity to estimate the value of Palm forwards, since options were traded on Palm shortly after the 5% spin-off:

Note that it is (almost) always positive, unlike the incorrectly computed stub that ignores the value of

lending fees embedded in Palm shares.

**The Remaining Puzzle**

Spatt admitted that it remains unknown why some investors held Palm shares without lending them, since the market must, in aggregate, hold the shares long. Perhaps these investors were naÃ¯ve, could not lend shares, and/or were overvaluing Palm shares. Whatever the case, the 3Com-Palm puzzle, if not fully resolved, is now focused on this simpler, new puzzle.

**Bibliography**

Cherkes, Martin, Charles M. Jones, and Chester S. Spatt, A Solution to the Palm-3Com Spin-Off Puzzles,

2013, available at https://www.q-group.org/wp-content/uploads/2014/11/Spatt-paper-Palm3Com20150218.pdf.

Lamont, Owen and Richard H. Thaler, 2003, Can the Stock Market Add and Subtract? Mispricing in Tech

Stock Carve-Outs, Journal of Political Economy 111, 227-268.

Mitchell, Mark, Todd Pulvino, and Erik Stafford, 2002, Limited Arbitrage in Equity Markets,” Journal of

Finance 57, 551-584.

Anna Scherbina, Associate Professor of Finance at the University of California, Davis, presented a lecture that contributed to our understanding of how information gets impacted into stock prices. She showed that it is possible, using Granger causality methodology, to identify, statistically, a collection of bellwether stocks that lead other followers. Most of the previous literature has focused on using ex-ante identifiable company characteristics to identify the leaders. These have included firm size, analyst coverage, supplier/customer relationships, strategic alliances, and merger prospects. In contrast, Scherbina focuses on situations where the information may be related to important news developments, may be transitory, and is difficult to identify, ex-ante.

Scherbina motivated her empirical analysis by discussing several examples of firm-level news that could be relevant for other firms: Texaco and an employee discrimination lawsuit in 1994-1996, Novartis and an Indian patent law case in 2012, John Wiley & Sons and the resale of U.S. items priced cheaper abroad in 2008-2013, and Worldcom’s earnings manipulation in 1999-2002. On average, she noted that there are about 218 important firm-level news items issued daily, and that only about 20% are related to financial information. Her empirical results looked at whether these information releases cause some firms to be market leaders — with information only gradually over time reflected in followers’s stock prices.

She identified a set of market leaders using the following Granger causality specification to run a series

of rolling one-year regressions for all possible pairs of stocks:

Having identified a set of Jt i leaders for stock i at time t, Scherbina aggregates all of the leaders’ signals:

where the weights (w) are either equal or value weights, and then examines whether the signals have outof-

sample forecasting ability. Ten decile portfolios are formed based on the composite leader signal. Portfolio 1 (10) contains the stocks that are predicted to perform the most poorly (the best). The results for the equally-weighted signal are in the following table; they indicate an out-of-sample alpha of 0.64% per month, based on the Fama French 4-factor alpha, from going long Portfolio 10 and short portfolio 1. Scherbina found that the leaders can be smaller stocks and they can belong to different industries than their followers.

Although the results are strongly significant over the entire sample period, Scherbina presented a graph (not reproduced) of cumulative monthly returns that showed the predictability at monthly frequencies has declined over time, and that the strategy may no longer be profitable. A similar graph (reproduced below) shows that some profitability remains when the strategy is executed on a weekly basis. The strategy requires high turnover; the break-even trading costs are about 45 bps for the equally-weighted portfolio strategy and about 27 bps for the value-weighted strategy-the same order of magnitude as the effective bid-ask spread for a typical stock.

In conclusion, Scherbina noted that it is possible to identify leaders based on Granger causality regressions. The existence of leaders is consistent with the limited attention of investors and the costs of acquiring information. Given the size of trading costs faced by most investors, prices appear to lie within no-arbitrage trading bounds.

__Bibliography__

Scherbina, Anna, and Bernd Schlusche, Cross-Firm Information Flows and the Predictability of Stock

Returns, Working Paper, available at https://www.q-group.org/wp-content/uploads/2014/11/QGroup_paper_Scherbina.pdf

Kent Smetters, Boettner Professor of Business Economics and Public Policy at the Wharton School, presented a new and “Sharper Ratio”! It is known that, when trading occurs at a higher frequency (e.g., daily) than the periods used to compute a Sharpe Ratio (e.g., monthly), a fund manager can use a dynamic strategy to artificially boost the Sharpe Ratio without any particular investment skills (e.g., Goetzmann,Ingersoll, Spiegel, and Welch, 2007). It is also known that certain strategies, such as a “selling putoption” strategy can generate elevated Sharpe Ratios in the absence of any manager skills. In both cases, the resulting distributions of manager returns are non-normal, violating a key assumption of the Sharpe Ratio. Another perspective on the problem is that these trading strategies alter higher moments of the return distribution in ways that can destroy the Sharpe Ratio’s usefulness as a performance evaluation tool.

Smetters motivated his talk with the following example: Consider the following baseline investment

strategy:

1. 50% of wealth invested in a hypothetical risky asset whose returns are generated from an i.i.d.

normal distribution, calibrated to have the same mean and standard deviation as the S&P1500

2. 50% in the risk-free asset

3. Portfolio rebalanced each week to 50/50

4. Sharpe Ratio and all return moments are measured on final annual return distribution

The resulting Sharpe Ratio is 0.62, with higher moments shown below:

So far, this is showing that the rebalance strategy-increasing risk when portfolio value drops, and decreasing when it rises-creates a slightly non-normal portfolio return distribution over a year (as shown by, for example, that skewness (M3) equals 0.232, rather than zero-its value when normally distributed.)

Now, let’s add, to this baseline strategy, a strategy of selling 10% out-of-the-money three-month put options on the index-with the number of options sold designed to generate 3%/year return (note that put options are a synthetic dynamic trading strategy that gets rebalanced every instant, again creating a problem for the Sharpe Ratio whose usefulness is predicated on a stationary underlying distribution):

Notice how the “Baseline + Puts” strategy is different from the “Baseline” strategy. The B+P strategy generates lower negative moments (M3, M5, M7, and M9) and higher positive moments-when rational investors will prefer exactly the opposite! Yet, its substantially increased Sharpe Ratio falsely signals that this is a superior investment.

As another example, suppose that we invest $1 in a 50/50 mix of the randomly generated “S&P 1500” plus riskfree asset. Each week, we rebalance according to a “Sharpe-Ratio optimal rule” derived by Smetters, and reflected in the below graph:

Simply put, the graph says to increase risk after losing money and decrease risk after making money. While this could be considered a “cheater’s graph,” a contrarian investing strategy follows this approach (buy when the market is down, sell when it is up). The result, compared with our above two strategies, is shown below:

Again, we have created an increased Sharpe Ratio, over the Baseline strategy, but at the expense of giving the investor worse higher moments.

One common way of dealing with this problem in practice is to use the Sharpe ratio to rank investment alternatives in combination with other measures of risk such as maximum drawdown, Sortino ratio etc. Smetters argues that while using multidimensional measures of risk sounds sophisticated, it lacks a theoretical foundation, often has to resolve conflicting conclusions, and that, ultimately, expected utility is the most compelling way to trade off risk and return.

Smetters motivated his “Sharper Ratio” by briefly reviewing past attempts to extend the Sharpe ratio. The prior literature typically started with a simple problem of maximizing the expected utility of an investor with unknown preferences. This literature showed that the first-order condition for utility maximization (first derivative with respect to allocation to stocks) can be written as an infinite Taylor Series expansion (which has a unique single root). Then, to produce a measure that ranks by the first N moments, you can truncate the Taylor Series expansion at N terms. Unfortunately, this literature stopped here, as there are many roots, some real and some complex (when the series is truncated at N terms), so it’s not clear what to do next!

Smetters’ ability at math becomes apparent next. He proposed an approach for selecting the correct root from the many roots noted above. The idea is that, if a power series (such as a Taylor Series) is truncated at N terms, then it has N roots. Then, under certain regularity conditions on the utility function and portfolio risk distribution (which are not too restrictive), the smallest (in absolute value) real root among the N roots converges to a unique root as we increase N to infinity. And, there is a finite value of N such that we can stop at this N and we will have our unique real root. This solves the problem noted above, and we can now use the N-term Taylor Series!

After demonstrating how the Taylor Series can be used to form a new “Sharper Ratio,” Smetters argued that the new ratio retains several attractive properties of the Sharpe Ratio, including that it is independent of leverage, but now properly accounts for higher moments of the portfolio return distribution. The only cost is that the Sharper Ratio depends on the risk tolerance of the investor (while Sharpe Ratio does not). Smetters argued that this should not be too burdensome, as financial advisors are charged with understanding the risk-aversion of their clients anyway!

__Bibliography__

Goetzmann, William, Jonathan Ingersoll, Matthew Spiegel, and Ivo Welch, 2007, Portfolio performance manipulation and manipulation-proof performance measures, Review of Financial Studies 20, 1503–1546.

Smetters, Kent, 2015, A Sharper Ratio, available at https://www.q-group.org/wp-content/uploads/2014/11/Smetters-paper-SharperRatio2-3-2014.