Hello. I need to write a three page review for the two articles attached. Each review should include a brief summary that include the major points of the article, as well as a summary of my own ideas/opinions on the topic discussed. I need both article reviews no later than 7pm EDT tonight.

If you have any questions, please let me know.

Thank you

Financial Analysts Journal Volume 70 Number 2 2014 CFA Institute The Low-Risk Anomaly: A Decomposition into Micro and Macro Effects Malcolm Baker, Brendan Bradley, and Ryan Taliaferro Low-risk stocks have offered a combination of relatively low risk and high returns. We decomposed the lowrisk anomaly into micro and macro components. The micro component comes from the selection of low-beta stocks. The macro component comes from the selection of low-beta countries or industries. Both parts contribute to the anomaly, with important implications for the construction of managed-volatility portfolios. I n an efficient market, investors earn higher returns only to the extent that they bear higher risk. Despite the intuitive appeal of a positive risk-return relationship, this pattern has been surprisingly hard to find in the data, dating at least to Black (1972). For example, sorting stocks by using measures of market beta or volatility shows just the opposite. Panel A of Figure 1 shows that from 1968 through 2012 in the US equity market, portfolios of low-risk stocks delivered on the promise of lower risk as expected but had surprisingly higher average returns. A dollar invested in the lowest-risk portfolio grew to $81.66, whereas a dollar invested in the highest-risk portfolio grew to only $9.76. A similar inverse relationship between risk and return from 1989 through 2012 in a sample of up to 31 developed equity markets can be seen in Panel B of Figure 1. A dollar invested in the lowest-risk portfolio of global equities grew to $7.23; a dollar invested in the highest-risk portfolio grew to only $1.20. This so-called low-risk anomaly suggests a very basic form of market inefficiency. Malcolm Baker is a professor of finance at Harvard Business School, research associate at the National Bureau of Economic Research, and senior consultant at Acadian Asset Management LLC, Boston. Brendan Bradley is director of portfolio management and Ryan Taliaferro is a portfolio manager at Acadian Asset Management LLC, Boston. Authors' note: The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research or Acadian Asset Management. The views should not be considered investment advice and do not constitute or form part of any offer to issue or sell, or any solicitation of any offer to subscribe or to purchase, shares, units or other interests in any particular investments. Shiller (2001) credited Paul Samuelson with the idea of separating market efficiency into two types. Micro efficiency concerns the relative pricing of individual stocks, whereas macro efficiency refers to the pricing of the market as a whole. Broadly speaking, inefficiencies can be examined at different levels of aggregation: at the individual stock level, at the industry level, at the country level, or, in some cases, at the global level. Samuelson (1998) conjectured that capital markets have \"come a long way, baby, in 200 years toward micro efficiency of markets: Black-Scholes option pricing, indexing of portfolio diversification, and so forth. But, there is no persuasive evidence, either from economic history or avant-garde theorizing, that macro market inefficiency is trending toward extinction\" (p. 36). At the heart of the difference is the fact that individual securities often have close substitutes. As Scholes (1972) showed, the availability of close substitutes facilitates low-risk micro arbitrage and pins down relative pricesif there are no practical limits on arbitrage. Industry and country portfolios have fewer close substitutes, and the equity market as a whole has none. So, individual stocks, in this view, are priced more efficiently relative to each other than they are in absolute terms. Of course, limits to arbitrage are real and substitutes are never perfect, so even Samuelson's hypothesis is one of relative, not absolute, efficiency. In the context of the low-risk anomaly, Baker, Bradley, and Wurgler (2011) emphasized the important constraint that long-only, fixed-benchmark mandates impose on micro arbitrage. Many institutional investors are judged not by total return relative to total risk but instead by active return relative to active risk, or benchmark tracking error. Such benchmark-oriented mandates discourage investment in low-risk stocks. Despite their low risk, these March/April 2014\twww.cfapubs.org FAJ_March-April_2014.indb 43 43 3/12/2014 9:32:38 PM Financial Analysts Journal Figure 1. \u0007 Cumulative Returns of Five Portfolios Formed on Stocks' Prior Betas: US Stocks (1968- 2012) and Developed-Market Stocks (1989-2012) A. Value of $1 Invested in 1968 Dollars 100 Bottom Quintile 10 Top Quintile 1 0.1 68 72 76 80 84 88 92 96 00 04 08 12 B. Value of $1 Invested in 1989 Dollars 100 10 Bottom Quintile 1 Top Quintile 0.1 88 92 96 00 04 08 12 Notes: We used quintile breakpoints at the beginning of each month to assign each stock in the CRSP database (United States, Panel A) and in the S&P Developed Broad Market Index (BMI) database (developed markets, Panel B) to one of five portfolios on the basis of its beta, estimated using the prior 60 (minimum 12) months (Panel A) or weeks (Panel B) of returns. We capitalization weighted the stocks in the resulting five portfolios, and we plotted the cumulative returns of investing US$1.00 in each. In Panel B, countries included for the full sample are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Hong Kong, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, Singapore, Spain, Sweden, Switzerland, the United Kingdom, and the United States; countries included for part of the sample are the Czech Republic, Greece, Hungary, Iceland, Israel, Luxembourg, Malaysia, Portugal, Slovenia, and South Korea. Betas are calculated with respect to the cap-weighted portfolio of US stocks using monthly returns (Panel A) and with respect to the cap-weighted portfolio of developed-market stocks using weekly returns (Panel B). All returns are measured in US dollars. Returns greater than 1,000% or less than -100% are excluded from the BMI database. Observation windows are January 1968 through December 2012 (Panel A) and November 1989 through December 2012 (Panel B). stocks become attractive relative to the tracking error they create only when their anticipated return exceeds the benchmark return in absolute terms. There are also limits on macro arbitrage. Market aggregates do not have close substitutes, and so macro arbitragein the usual sense of simultaneously buying low and selling fundamentally similar securities highis largely infeasible. Standard institutional mandates and risk management practices also play a role here, typically limiting the size of benchmark-relative country or industry exposures or eliminating them entirely through narrow mandates that identify a single-country index as the benchmark return. French and Poterba (1991) and 44\twww.cfapubs.org\b FAJ_March-April_2014.indb 44 Ahearne, Griever, and Warnock (2004) documented a home bias, for example, showing that individuals often do not invest across borders. One way of examining the relative efficiency of markets at different levels of aggregation is to aggregate or disaggregate a known anomaly into its micro and macro components. An important limitation of this approach is that anomalies change their meaning at different levels of aggregation. Ratios of fundamental to market value capture misvaluation at the individual stock level, as in Basu (1977). But Lewellen (1999) found little contribution from the industry level, where differences in valuation ratios might reflect varying approaches to capitalizing 2014 CFA Institute 3/12/2014 9:32:39 PM The Low-Risk Anomaly investment or reporting earnings. Fama and French (1998) found that predictive power reappears at even higher levels of aggregation in cross-country regressions. Likewise, Kothari and Shanken (1997) found that time-series comparisons of valuation ratios for equity markets as a whole are also useful. Unlike value, beta retains the same meaning at the stock level and in country and industry portfolios, though it loses some of its variability. Its usefulness runs out for the market as a whole, where it is 1.0 by definition. Another limitation of aggregation is that econometric tests typically lose power to identify inefficiency. So, it is important to consider economic and statistical significance. Discussionoffindings. Inthespiritof Samuelson, we decomposed the low-risk anomaly into its micro and macro components. The pattern of low risk and high return can, in principle, come from either the macro selection of lower-risk countries and industries or the micro selection of low-risk stocks within those countries and industries. We separated the two effects by forming long-short portfolios of stocks and held constant ex ante country- or industry-level risk to examine stock selection. Next, we held constant ex ante stock-level risk and examined country and industry selection. What we found in a sample of 29 US industries and up to 31 developed countries is that both micro selection and macro selection contribute to the low-risk anomaly, albeit for two surprisingly different reasons. The micro selection of stocks leads to a significant reduction in risk, with only a modest difference in return. Even holding constant country- or industry-level risk, there is ample opportunity to form lower-risk portfolios through stock selection, and these portfolios do not suffer lower returns. High-risk stocks can be distinctly identified within the utility industry or in Japan, for example, but they have similar raw returns on average when compared with low-risk stocks in the same industry or country grouping. This evidence supports the notion of limits to micro arbitrage in low-risk stocks that come from traditional fixed-benchmark mandates. Meanwhile, the macro selection of countries in particular leads to increases in return, with only modest differences in risk. Countries that we identify as high risk ex ante are only modestly higher risk going forward, but they have distinctly lower returns. This evidence supports the limits to macro arbitrage across countries and industries. Other researchers have examined the question of micro versus macro efficiency. Jung and Shiller (2005) showed that dividend yields predict the growth in dividends at the firm level but not at higher levels of aggregation. Intuitively, a higher dividend yield should be associated with lower growth in dividends, and across firms, this is indeed true, suggesting a degree of micro efficiency. Somewhat pathologically, higher aggregate dividend yields predict higher, not lower, dividend growth, suggesting macro inefficiency. This phenomenon parallels the findings in the more indepth studies by Campbell (1991) and Vuolteenaho (2002) of variance decompositions of stock returns using valuation ratios. Lamont and Stein (2006) extended this logic to corporate finance. Such corporate financing activities as new issuance and stock-financed mergers, which are in part designed to take advantage of market inefficiency, display greater sensitivity to aggregate movements in value than they do to individual stock price changes. Decomposing the low-risk anomaly is a particularly appealing new avenue because beta retains its meaning in country and industry aggregations. Unlike beta, value and momentum, for example, have no ex ante risk-based theory of return predictability. So, if value, for example, has greater predictive power within industries than across industries, little is revealed about micro versus macro efficiency. It may simply be differences in accounting conventions across industries that make relative value comparisons with financial statement data easier within industries. Our results on the decomposition of the lowrisk anomaly into micro and macro effects have investment implications for plan sponsors and individuals alike. For individuals, they suggest that trying to exploit the mispricing through industry, sector, or exchange-traded funds will get them only so far. Although such a strategy can gain exposure to the macro effects, it cannot exploit the considerable risk reduction available in micro stock selection. For institutional investors and plan sponsors, our results suggest that perfunctory approaches to risk modeling and overly constrained mandates will not fully appreciate the benefits of the macro effects. Broad mandates with loose constraints and thoughtful techniques for modeling risk will do the most to exploit the low-risk anomaly. The Low-Risk Anomaly A preliminary step is confirming that the low-risk anomaly is present in two sources of stock return data that are convenient for our decomposition. We sorted stocks into quintiles according to ex ante market beta and formed five capitalization-weighted, buy-and-hold portfolios. Theory predicts a positive relationship between risk and return. Under the Sharpe-Lintner capital asset pricing model (CAPM), the relationship is linear and the slope between contemporaneously measured market March/April 2014\twww.cfapubs.org FAJ_March-April_2014.indb 45 45 3/12/2014 9:32:39 PM Financial Analysts Journal beta and returns is equal to the overall market risk premium. Fortunately, the low-risk anomaly does not hinge on elaborate estimations because the empirical slope has the wrong sign. For our analysis of industries within the United States, we used data from the Center for Research in Security Prices (CRSP). From CRSP, we collected the monthly returns of all US shares for the period January 1963 through December 2012. We also collected the corresponding data on market capitalizations to form cap-weighted portfolios. For each month and each stock, we estimated a market beta using the previous 60 months of excess returns. We included stocks that had fewer than 60 prior monthly returns if they had a history of at least 12 months. For our analysis of countries across the developed world, we used data from the S&P Developed Broad Market Index (BMI). From the BMI, we collected monthly and weekly returns and market capitalizations for common stocks for the period July 1989 through December 2012. The raw dataset contains a very small number of suspicious returnsthose less than -100% and greater than 1,000%and we excluded these observations from our analysis. Market betas were estimated using the excess return of the capitalization-weighted aggregate of all stocks in the sample as the market portfolio. Returns are in US dollars, and the riskfree rate is the US Treasury bill rate from Kenneth French's website.1 For each month and each stock, we estimated a market beta using the previous 60 weeks of returns. Stocks that had fewer than 60 prior weekly returns were included if they had a history of at least 12 weeks. We opted for a more dynamic estimation of betas in the BMI. The overall sample is shorter, and a five-year period to estimate initial betas is more costly. Neither monthly nor weekly country betas perform particularly well in predicting future country portfolio betas, but weekly measures provide a marginal improvement. Figure 1 depicts the low-risk anomaly graphically. In the CRSP data in Panel A, a $1.00 investment in the low-risk quintile portfolio in 1968 compounds to $81.66 by the end of 2012. A $1.00 investment in the high-risk quintile compounds to only $9.76. Over the shorter history of the BMI data, a $1.00 investment in 1989 compounds to $7.23 in the low-risk quintile and $1.20 in the highrisk quintile. To understand the statistical significance of these results, we performed a simple test in Table 1. We regressed the returns of each of the resulting five portfolios, measured in excess of US Treasury bill returns, on the aggregate market excess return to find the portfolio's full-period ex post market 46\twww.cfapubs.org\b FAJ_March-April_2014.indb 46 beta and CAPM alpha. The first and third columns in Table 1 report the betas and alphas, respectively, of five CRSP portfolios. The second and fourth columns in Table 1 report the same calculations for the BMI data. For example, the portfolio formed from the CRSP stocks with the lowest betas realized a beta of 0.59 over the full period (1968-2012). The portfolio formed from the CRSP stocks with the highest betas realized a beta of 1.61. The difference between these betas, -1.02, is both large and significant, with a t-statistic of -23.99. These results suggest that betas estimated from past returns are strong predictors of future betas. More importantly, the third column in Table 1 reports the alphas for five CRSP beta-sorted portfolios. The difference in risk-adjusted performance is stark. The least risky portfolio, formed from the lowest-beta stocks, has an alpha of 2.27% and a t-statistic of 2.16, whereas the riskiest portfolio, formed from the highest-beta stocks, has an alpha of -4.49% and a t-statistic of -2.70. The difference between these alphas, 6.76 percentage points (pps), is economically large and statistically significant, with a t-statistic of 2.84. It is this last result that summarizes the low-risk anomaly: Low-risk stocks outperform high-risk stocks on a risk-adjusted basis. Despite a shorter history, the low-risk anomaly appears to be stronger in the BMI data. The fourth column in Table 1 reports the alphas for five BMI beta-sorted portfolios. The difference between the alphas, 9.24 pps, is larger and has a t-statistic of 2.71. This test is not quite independent of the low-risk anomaly because the samples overlap in time and in securities, with the United States included in both. However, when we exclude the two largest countries, the United States and Japan, we obtain similar results both here and in our decompositions. Our aim is to use data of the highest quality and with the longest history for separate industry and country decompositions. We sacrifice cross-country variation to double the history of our industry decomposition, and we sacrifice history to have a stable cross-country decomposition. These preliminary results align with the literature, which dates to Black (1972); Black, Jensen, and Scholes (1972); and Haugen and Heins (1975). More recently, Fama and French (1992), Falkenstein (1994), and others have observed a flat or even negative empirical relationship between risk and return. Ang, Hodrick, Xing, and Zhang (2006, 2009) focused more on volatility sorts, pointing out the poor performance of the highest-volatility deciles. Bali, Cakici, and Whitelaw (2011) developed a related measure of 2014 CFA Institute 3/12/2014 9:32:40 PM The Low-Risk Anomaly Table 1. \u0007 CAPM Betas and Annualized Alphas for Portfolios Formed on Stocks' Prior Betas CAPM (% per year) CAPM CRSP Stock A. Point estimates Low 2 3 4 High Low - High B. t-Statistics Low 2 3 4 High Low - High BMI Stock CRSP Stock BMI Stock 0.59 0.76 0.98 1.24 1.61 -1.02 0.47 0.69 0.88 1.16 1.69 -1.22 2.27 2.10 0.28 -1.52 -4.49 6.76 3.74 3.60 2.01 -1.32 -5.50 9.24 31.24 49.56 72.75 85.93 54.09 -23.99 18.42 30.98 46.05 69.22 41.95 -19.89 2.16 2.45 0.37 -1.89 -2.70 2.84 2.62 2.92 1.89 -1.42 -2.45 2.71 Notes: The data are from CRSP for January 1968 through December 2012 and from the S&P BMI for November 1989 through December 2012. We used quintile breakpoints at the beginning of each month to assign each stock to a portfolio based on its own beta. Betas in the CRSP sample were estimated using the prior 60 months (minimum 12 months) of returns. Betas in the BMI sample were estimated using the prior 60 weeks (minimum 12 weeks) of returns, where all returns are measured in US dollars. For each set of breakpoints, we capitalization weighted the stocks in the resulting five portfolios, and we regressed each portfolio's monthly excess returns on a constant and the market excess returns to find full-period ex post CAPM betas (left columns) and alphas (right columns). To compute excess returns, we used the US government one-month T-bill rate as the risk-free rate. Panel A reports point estimates for these regressions, and Panel B reports corresponding t-statistics. Countries included for the full sample are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Hong Kong, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, Singapore, Spain, Sweden, Switzerland, the United Kingdom, and the United States; countries included for part of the sample are the Czech Republic, Greece, Hungary, Iceland, Israel, Luxembourg, Malaysia, Portugal, Slovenia, and South Korea. We excluded returns greater than 1,000% and less than -100%. For CRSP, the market portfolio is from Kenneth French's website. For the BMI, the market portfolio is the cap-weighted portfolio of all included stocks. lottery-like risk. Hong and Sraer (2012) incorporated speculative motives to trade into an asset pricing model to explain the low-risk anomaly, whereas Frazzini and Pedersen (2010) focused on the implications of leverage and margining and extended this focus to other asset classes. There is a rough risk-return trade-off across long histories of asset class returns, where stocks outperform bonds, for example. But within asset classes, the lowest-risk portfolios have higher Sharpe ratios. Baker and Haugen (2012) and Blitz, Pang, and van Vliet (2012) documented the outperformance of low-risk stocks in various markets throughout the developed and emerging world. None of these authors focused on the decomposition of the anomaly into micro and macro effects. Asness, Frazzini, and Pedersen (2013) applied a different methodology to examine the contributions to the low-risk anomaly of the industry and stock selection pieces. Their results are consistent with our own decomposition at the industry level. Explanations have emphasized a combination of behavioral demand and the limits to arbitrage, including limited borrowing capacity and the delegation of stock selection, to explain the lowrisk anomaly. Baker et al. (2011) surveyed three behavioral explanations: lottery preferences, representativeness, and overconfidence. We omit a detailed review of these explanations here to save space. In terms of constraints, Black (1972) and Frazzini and Pedersen (2010) emphasized leverage. Karceski (2002) and Baker et al. (2011) focused on the effects of intermediation. Whereas Karceski described the interaction between highbeta strategies and the capture of mutual fund flows, Baker et al. derived the implications of fixed-benchmark strategies for low-risk stocks using the framework in Brennan (1993). It is worth noting that these stories suggest both the overvaluation of high-risk stocks and the undervaluation of low-risk stocks. Our focus here is not to disentangle the absolute mispricing but, March/April 2014\twww.cfapubs.org FAJ_March-April_2014.indb 47 47 3/12/2014 9:32:40 PM Financial Analysts Journal rather, to examine the relative mispricing in Table 1. Assuming that the behavioral demand applies to both individual stocks and market aggregates, a decomposition of the low-risk anomaly can, in principle, help us understand the extent of the limits to micro and macro arbitrage, as well as the role of fixed benchmarks. Variations on the Low-Risk Anomaly There are three important footnotes to our analysis thus far. The first involves the rate of turnover and liquidity in the low-risk portfolios. The second involves the benchmark model used for computing risk-adjusted returns, or alphas. The third is the measure of risk itself, which has varied in the literature. Beta Turnover and Liquidity. Some measures of risk require high-frequency rebalancing. For example, Ang et al. (2006, 2009) and Bali et al. (2011) measured idiosyncratic risk over relatively short horizons. In contrast, the results in Table 1 do not require much turnover. For the CRSP sample, this finding is intuitive. The betas were computed using 60 months of data, so changes in beta estimates from month to month are small. The probability of staying in the first or fifth quintile from one month to the next is greater than 95%, and the probability of moving out of the top or bottom two quintiles from the first or fifth quintile is less than 0.5%. Moreover, lagged betas show a similar ability to predict monthly returns, so turnover is of limited value. Table 2 repeats the analysis in the third column of Table 1 using betas lagged up to 12 months. Economic and statistical significance remains almost unchanged. In terms of liquidity, Baker et al. (2011) showed that the pattern in Figure 1 appears more robust when attention is restricted to the largest 1,000 stocks in the CRSP data. Although we have not modeled transaction costs as in Li, Sullivan, and Garcia-Feijo (2014), the combination of these two factslow turnover and the existence of the anomaly in large-cap stockssuggests this particular low-beta strategy was implementable over this time period. Benchmark Models. There are several possibilities in the literature for benchmarking returns. We chose to focus on CAPM regressions for two reasons. The first is that we can parsimoniously show the impact on both risk and alpha in double sorts. With three or four measures of risk, doing so becomes complicated. We will show how firm-, industry-, and country-level information affects portfolio risk reduction and risk-adjusted returns, which works nicely with CAPM betas. The second, and related, reason is that beta is a theoretically motivated measure of risk. Low-beta stocks and portfolios of low-beta stocks are less risky in the sense that their marginal contribution to the overall volatility of most diversified portfolios is negative. Size, value, and momentum are other anomalies that were discovered because of their return properties not because of any clear risk-based theory. Even if size and value can also be considered anomalous, it is still useful to ask whether beta sorts represent an anomaly different from those of value and size. As it turns out, low-risk alphas are still statistically significant with controls for value and size. The alpha in Table 1 retains its statistical Table 2. \u0007 CAPM Annualized Alphas for Portfolios Formed on Stocks' Prior Betas CRSP Stock CAPM (% per year) CRSP t - 2 CRSP t - 3 Stock Stock CRSP t - 12 Stock A. Point estimates Low 2 3 4 High Low - High 2.27 2.10 0.28 -1.52 -4.49 6.76 2.29 1.89 0.27 -1.34 -4.81 7.10 1.82 1.66 0.26 -1.06 -4.65 6.48 2.00 1.87 0.05 -0.98 -3.57 5.58 B. t-Statistics Low 2 3 4 High Low - High 2.16 2.45 0.37 -1.89 -2.70 2.84 2.20 2.10 0.37 -1.69 -2.98 3.05 1.72 1.95 0.34 -1.36 -2.86 2.76 1.92 2.22 0.06 -1.32 -2.36 2.52 Note: This table shows our results when we repeated the analysis in Table 1 using lagged values of beta. 48\twww.cfapubs.org\b FAJ_March-April_2014.indb 48 2014 CFA Institute 3/12/2014 9:32:41 PM The Low-Risk Anomaly significance, but the point estimate drops from 6.76 to 4.08. This finding tells us that there is incremental information in beta that is not fully contained in value and also that some of the alpha in lowrisk strategies can be captured through value. This is not very surprising; any explanation where low-risk stocks are undervalued predicts that crude value measures absorb some of the low-risk anomaly. Like value, momentum helps to explain some of the performance of low-beta strategies. Momentum alone does not eliminate the low-beta outperformance, but it does somewhat reduce the economic and statistical power of the outperformance. However, there is even less reason to include momentum in the benchmark. The correlations between value and beta and between size and beta are stable, but the connection between momentum and beta is not. With large market moves, momentum portfolios can inherit either high beta risk or low beta risk. Occasionally, the extreme quintiles overlap, and sometimes, they contain a disjoint set of stocks. So, the low-risk anomaly has no routine, theoretical, and positive exposure to momentum as a risk factor. Rather, it is momentum that occasionally has low or high market risk. Daniel and Moskowitz (2013) showed that momentum crashes have occurred disproportionately following major market declines, when past losers are high beta and past winners are low beta. In these periods, low-beta stocks have also underperformed high-beta stocks. This strikes us as an interesting feature of momentum but not a reason to change the benchmark. Measures of Risk. Beta is one of several measures of risk that are associated with a broad lowrisk anomaly. The reason we focus on beta is that any portfolio beta, including betas of industry or country portfolios, is a simple weighted-average combination of the individual security betas contained in that portfolio, at their portfolio weights. The same cannot be said of a portfolio's idiosyncratic volatility. The full covariance matrix of the individual securities is needed to go from security-level risk to portfolio-level risk. Nonetheless, a decomposition of idiosyncratic risk, given its ability to predict stock-level returns, is an interesting robustness check. The main choices in measuring volatility are the benchmark and the time horizon. We focused on a low-frequency measure of idiosyncratic volatility that draws from the CAPM regressions described previously. Although we did not match the approach in Ang et al. (2006, 2009), the resulting portfolios are more stable, with lower transaction costs in implementation, and the interpretation is simpler because the estimation window and the regression specification are identical. A Decomposition of Micro and Macro Effects in 29 Industries We begin with a decomposition of the low-risk anomaly into separate industry- and stock-level effects. It is worth keeping in mind the statistical significance of the basic anomaly. There are economically large differences in both risk and riskadjusted returns, but the statistical significance of the return differences is more modest, which suggests that we have somewhat less power to decompose these effects definitively, and we focused on economic effects more than statistical significance. We grouped firms into industries using their primary SIC memberships, as identified by CRSP. Because four-digit SIC codes divide the sample into groups that are sometimes too small to obtain robust estimates of risk and return for, we used Kenneth French's somewhat broader industry groupings that build on Fama and French (1997). Our industry definitions are subjective to a degree. For example, on Kenneth French's website, there are definitions that group stocks into 5, 10, 12, 17, 30, 38, 48, and 49 industries. In the analysis that follows, we used the definitions for the 30 industries, dropping the \"Other\" category to net 29 industries, and we have checked that the results are not materially different when using the 12 industry definitions. Two considerations motivated our use of the 30 industry definitions. First, the 12 industry definitions leave a substantial subset of stocks categorized as \"Other,\" whereas the 30 industry definitions assign a much smaller subset to this classification. Second, our methodology groups stocks into quintiles based on industry betas, and it seemed unnecessarily coarse to use 12 industry definitions to assign stocks to five quintiles. When there are more than 30 industries, however, there are fewer stocks per industry and the potential problems of using fourdigit SIC codes emerge, so we did not push beyond 30 industries into finer industry definitions. Using fewer industry groupings did not necessarily tilt us in favor of finding a more robust micro effect. What matters is the link between estimated and realized industry betas, and the robustness of this relationship initially increases with the number of industries but eventually decreases as the estimation error associated with thin cells becomes relevant. TheLow-RiskAnomalyacrossIndustries. Table 3 is a prelude to the decomposition of the low-risk anomaly, repeating the analysis in the first and third columns of Table 1 with industry betas in place of stock betas. A stock's industry beta is the capitalization-weighted average of the betas of the stocks within the industry; on each estimation date, the industry beta is the same for each stock March/April 2014\twww.cfapubs.org FAJ_March-April_2014.indb 49 49 3/12/2014 9:32:41 PM Financial Analysts Journal in a given industry. For each month, we used quintile breakpoints to assign an approximately equal number of stocks to five portfolios using industry betas. All stocks in a given industry were assigned to a single quintile portfolio. We again capitalization weighted the stocks in each portfolio to find portfolio returns, and we estimated market betas and CAPM alphas. The first column suggests that historical, ex ante industry betas are able to predict ex post realized stock betas, but not as well as historical stock betas do. In this column, the difference between the ex post betas of the high and low portfolios is -0.60, with a t-statistic of -16.69roughly two-thirds of the spread achieved using stock betas in Table 1. Similarly, compared with the difference in ex post alphas using stocks' own betas, the second column reports a smaller and marginally significant difference in ex post alphas of 3.65 pps, with a t-statistic of 1.80. In short, using only information about industry betas and no stock-level information delivers a reasonable fraction of the risk reduction and riskadjusted return improvement of stock-level sorts. Table 3. \u0007 CAPM Betas and Annualized Alphas for Portfolios Formed on Prior 60-Month Industry Betas: United States, January 1968-December 2012 CAPM Industry A. Point estimates Low 2 3 4 High Low - High B. t-Statistics Low 2 3 4 High Low - High CAPM (% per year) Industry 0.71 0.92 0.97 1.07 1.31 -0.60 2.25 2.50 -1.33 -0.40 -1.39 3.65 37.29 52.91 54.38 52.09 58.20 -16.69 2.11 2.58 -1.34 -0.35 -1.10 1.80 DecomposingtheLow-RiskAnomaly betwen Stocks and Industries.We now turn e to the main question: To what degree do macro effects contribute to the low-risk anomaly? In this case, industries represent the macro dimension of the anomaly. To understand the separate contributions of industry selection and stock selection within industries, we combined the independent 50\twww.cfapubs.org\b FAJ_March-April_2014.indb 50 quintile sorts of Table 1 and Table 3 into a doublesort methodology. For each month, we used independent quintile breakpoints to assign each stock to a portfolio based on its own beta and its industry beta. We capitalization weighted the stocks in the resulting 25 portfolios. We are interested in the contribution of stocklevel risk measures once industry risk has been taken into account and in the contribution of industry-level risk measures once stock-level risk has been taken into account. We can determine these contributions only to the extent that these sorts do not lead to the same exact division of the CRSP sample. Figure 2 plots the fraction of stocks in each of the resulting 25 portfolios in Panel A and the fraction of market capitalization in Panel B. If these sorts led to the same division of the CRSP sample, all the firms and the market capitalization would lie on the diagonal. The plots show that the two effects overlap considerably, especially when measured in terms of market capitalization. This finding is intuitive; large-cap firms are likely to have measures of risk that are closer to the capweighted industry beta. The plots also show that we have some ability to separate the two effects. As a side note, we also considered dependent sorts, in addition to the independent sorts described next. We first sorted stocks into quintiles using industry beta. Within each quintile, we then sorted using stock-level beta. Doing so resulted in an equal number of stocks in each of the 25 portfolios. Our findings are largely similar using this approach, for both the industry and the country analyses. For each of the 25 independent double-sorted portfolios, we estimated an ex post beta and CAPM alpha over the full period. Table 4 reports results. The left columns report betas, and the right columns report alphas. In the left columns, betas are widely dispersed and tend to increase from top to bottom and from left to right. For example, the first data column reports the betas of portfolios formed from stocks in low-beta industries: Reading down the column, portfolio betas within this subsample increase monotonically from 0.54 to 1.48. Similarly, the fifth data column reports betas of portfolios formed from stocks in high-beta industries, and the range of portfolio betas within this subsample is also large, increasing from 0.87 to 1.64. In contrast, rows display much less variation in portfolio betas. For example, the \"Low\" row, which consists of portfolios formed from stocks with the lowest betas, shows that betas range from 0.54 (lowest-beta industry) to 0.87 (highest-beta industry). Among high-risk stocks in the \"High\" row, the industry effect is even smaller. In this subsample, stocks in the highest-beta industries have an average 2014 CFA Institute 3/12/2014 9:32:42 PM The Low-Risk Anomaly Figure 2. \u0007 Distribution of Stocks across Portfolios Formed Jointly on Stocks' Betas and Their Industry Betas: United States, January 1968- December 2012 A. Average Distribution of Stocks across Double-Sorted Portfolios Average Fraction of Stocks 0.10 0.05 0 High Low 4 2 3 3 Stock Beta 2 4 High Industry Beta Low B. Average Distribution of Market Capitalization across Double-Sorted Portfolios Average Fraction of Market Value 0.10 0.05 0 High Low 4 2 3 3 Stock Beta 2 4 High Low Industry Beta Note: Vertical bars denote the time-series average fraction of stocks (Panel A) or fraction of market capitalization (Panel B) in each portfolio. beta of 1.64, compared with an average beta of 1.48 for stocks in the lowest-beta industries. The statistics at the bottom of Panel A summarize these patterns. The first is a pure industry effect, measured as the average of the differences between high-beta and low-beta industries, controlling for stock-level risk. This is the average of the differences reported in the \"Low - High\" column. The second is a pure stock effect, measured as the average of the differences between high-beta and low-beta stocks, March/April 2014\twww.cfapubs.org FAJ_March-April_2014.indb 51 51 3/12/2014 9:32:43 PM Financial Analysts Journal Table 4. \u0007 CAPM Betas and Annualized Alphas for Portfolios Formed Jointly on Prior 60-Month Betas and Prior 60-Month Industry Betas: United States, January 1968-December 2012 CAPM (% per year) CAPM Industry Stock Low A. Point estimates Low 0.54 2 0.70 3 0.88 4 1.24 High 1.48 Low - High -0.94 Industry High Low - High Low 2 3 4 High Low - High -0.34 -0.23 -0.22 -0.05 -0.16 2.71 2.91 0.57 0.59 0.92 3.87 3.31 3.21 0.89 -2.32 -2.17 -0.14 -0.80 -2.30 -7.88 -1.57 -1.09 1.37 -1.29 -3.71 1.79 4.28 0.11 -1.51 -4.61 0.92 -1.37 0.47 2.10 5.54 1.78 6.19 5.71 2.14 6.40 2 3 4 0.71 0.80 0.96 1.15 1.46 0.72 0.78 0.97 1.21 1.48 0.68 0.82 0.99 1.20 1.55 0.87 0.93 1.11 1.29 1.64 -0.75 -0.76 -0.88 -0.76 Industry effect (average of Low - High column) -0.20 1.53 Within-industry effect (average of Low - High row) -0.82 4.44 Total effect 5.97 B. t-Statistics Low 2 3 4 High 20.34 29.42 36.16 36.67 27.82 25.61 36.87 44.38 43.82 34.35 20.45 31.11 42.95 49.85 40.26 19.27 26.78 40.56 44.76 40.51 21.31 33.06 48.49 52.77 41.85 Low - High -16.54 -15.77 -16.15 -17.72 -14.56 Industry effect (average of Low - High column) Within-industry effect (average of Low - High row) -6.78 -6.23 -7.04 -1.26 -2.61 FAJ_March-April_2014.indb 52 2.51 2.73 2.64 0.61 -0.98 -1.11 -0.10 -0.63 -1.70 -3.83 -0.8 -0.63 1.01 -0.86 -1.73 0.78 2.71 0.08 -1.10 -2.11 0.56 2.31 2.16 0.77 0.33 -0.66 0.27 0.91 1.66 2.19 -6.71 0.92 -24.93 2.42 controlling for industry risk. This is the average of the differences reported in the \"Low - High\" row. Our ability to find low-risk portfolios through the selection of low-risk stocks within industries, at a -0.82 reduction in beta, is about four times as large as our ability to find low-risk portfolios through the selection of low-risk industries within stock risk groups. These results are consistent with stock-level beta, as estimated from historical data, being a much better predictor of future beta than industry beta is. However, both a stock-level historical beta and each stock's industry-level historical beta, each relative to the other, have incremental predictive power for future beta. When computing the t-statistics in Table 4, we took into account the empirical covariance among the 25 double-sorted portfolios. Specifically, we used a seemingly unrelated regression model that allows regression errors to be contemporaneously correlated among the 25 regressions but that assumes zero autocorrelations or cross-autocorrelations. Measures of risk and differences in risk are highly statistically significant. Differences in return are harder to detect in the data. The right panel of Table 4 reports alphas for the double-sorted portfolios. Alphas tend to decrease from top to bottom and from left to right. The 52\twww.cfapubs.org\b 1.84 2.18 0.42 0.31 0.31 \"Low - High\" column reports differences in alphas between stocks in low-beta and high-beta industries, controlling for the stocks' own betas. The first summary statistic reports the average of these differences, a pure industry effect of 1.53%. Similarly, the \"Low - High\" row reports the differences in alphas between low-beta stocks and high-beta stocks, controlling for industry betas. The second summary statistic reports the average of these differences, a pure stock effect of 4.44%. Historically, both industry selection and stock selection have made material contributions to the outperformance of low-risk investing strategies. We cannot rule out that the industry effect happened by chance, though our power to reject this null hypothesis is limited. The beta and alpha estimates in Table 4 suggest that pure stock alpha (the micro effect) and pure industry alpha (the macro effect) arise for different reasons. The pure industry alpha of 1.53% is present in spite of a small cross-industry beta difference of -0.20. By implication, the industry alpha obtains as a consequence of cross-industry portfolios exhibiting differences in returns, with modest differences in risk. In contrast, the pure stock alpha of 4.44% is present alongside a dramatic difference in withinindustry beta of -0.82. By implication, the stock alpha obtains as a consequence of within-industry 2014 CFA Institute 3/12/2014 9:32:44 PM The Low-Risk Anomaly portfolios exhibiting material differences in risk, with a more limited difference in returns. A Decomposition of Micro and Macro Effects in Up to 31 Developed-Country Markets Another way to slice the data is geographically. Just as some industries, such as utilities and health care, are less sensitive to global equity markets, some regions and countries are thought to have different risk properties. Japan, for example, has a low correlation with global equity markets in its recent history. Next, we turn to decomposing the low-risk anomaly into country and stock effects. The sample of developed markets in the BMI includes Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Hong Kong, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, Singapore, Spain, Sweden, Switzerland, the United Kingdom, and the United States. In addition, several more countries are included or excluded at some point during the sample. To avoid any lookahead bias, we followed S&P's point-in-time decisions and included these countries in the developedmarket sample whenever S&P labeled them as such. These countries are the Czech Republic, Greece, Hungary, Iceland, Israel, Luxembourg, Malaysia, Portugal, Slovenia, and South Korea. For each month and for each stock, we estimated a stock beta with respect to the global equity market using the previous 60 weeks of returns, measured from Wednesday to Wednesday. Stocks with a history shorter than 60 weeks were included if they had at least 12 weeks of returns. We defined the global equity market as the capitalization-weighted average return of the entire BMI developed-market sample. We also measured an analogous country beta. Each stock was mapped to a country, and the stock's country beta is the capitalization-weighted average of the betas of each stock within the country: On each estimation date, the country beta is the same for each stock in a given country. The Low-Risk Anomaly across Countries.We repeated the industry analysis for countries in the BMI sample. As before, Table 5 is a prelude to the decomposition of the low-risk anomaly, repeating the analysis in the second and fourth columns of Table 1 with country betas in place of stock betas. The first column suggests that historical, ex ante country betas are able to predict ex post realized stock betas but, like industry betas, not as well as historical stock betas do. In this column, the difference between the ex post betas of the high and low portfolios is -0.55, with a t-statistic of -8.78roughly half the spread achieved using stock betas in Table 1. The alpha differences are somewhat larger than those in the industry sorts. The second column shows a difference in ex post alphas that is smaller than in Table 1 but still reasonably large (6.86 pps, with a t-statistic of 1.97). In short, using only information about country betas and no stock-level information delivers roughly half the risk reduction and about two-thirds of the risk-adjusted return improvement of stock-level sorts. This combination of a substantial risk-adjusted return difference and lower risk reduction is a preview of the decomposition. As we will show, unlike the industry analysis, there is something extra in the country beta sorts, in terms of total returns, that is not captured in the stock beta sorts of Table 1. Table 5. \u0007 CAPM Betas and Annualized Alphas for Portfolios Formed on Prior 60Week Country Betas: Developed Markets, November 1989-December 2012 CAPM CAPM (% per year) Country Country A. Point estimates Low 2 3 4 High Low - High 0.77 0.86 0.99 1.12 1.32 -0.55 5.87 4.77 0.78 -1.49 -0.99 6.86 B. t-Statistics Low 2 3 4 High Low - High 21.71 27.05 31.81 33.49 30.15 -8.78 2.97 2.68 0.45 -0.80 -0.41 1.97 DecomposingtheLow-RiskAnomaly between Stocks and Countries.We next examined to what degree macro effects contribute to the low-risk anomaly, but in this case, countries represent the macro dimension of the anomaly. To understand the separate contributions of country selection and stock selection within countries, we repeated the independent quintile sorts of Table 4 in Table 6. For each month, we used independent quintile breakpoints to assign each stock to a portfolio based on its own beta and its country beta. We capitalization weighted the stocks in the resulting 25 portfolios. We are interested in the contribution of stocklevel risk measures once country risk has been taken into account and in the contribution of countrylevel risk measures once stock-level risk has been March/April 2014\twww.cfapubs.org FAJ_March-April_2014.indb 53 53 3/12/2014 9:32:44 PM Financial Analysts Journal Table 6. \u0007 CAPM Betas and Annualized Alphas for Portfolios Formed Jointly on Prior 60-Week Betas and Prior 60-Week Country Betas: Developed Markets, November 1989-December 2012 CAPM (% per year) CAPM Country Stock Low A. Point estimates Low 0.52 2 0.74 3 0.97 4 1.17 High 1.62 Low - High -1.09 Country High Low - High Low 2 3 4 High Low - High -0.13 -0.04 0.03 0.08 -0.06 5.49 7.55 2.68 2.60 0.87 6.27 6.77 6.94 2.00 -0.39 1.97 3.51 2.29 0.11 -3.73 0.16 0.71 0.60 -2.35 -8.71 -3.62 -2.94 0.09 -0.69 -4.74 9.11 10.48 2.59 3.29 5.61 4.62 6.66 5.7 8.88 1.11 2 3 4 0.52 0.71 0.88 1.1 1.59 0.64 0.79 0.96 1.14 1.62 0.62 0.83 0.99 1.19 1.51 0.65 0.78 0.93 1.09 1.67 -1.07 -0.98 -0.89 -1.02 Country effect (average of Low - High column) -0.02 Within-country effect (average of Low - High row) 6.22 -1.01 5.40 Total effect 11.61 B. t-Statistics Low 2 3 4 High 14.35 18.87 21.58 19.17 15.39 12.29 19.40 21.79 22.63 21.62 13.27 21.71 28.00 32.45 26.71 12.62 18.83 25.07 30.56 23.22 10.41 15.77 19.94 23.93 27.52 Low - High -10.08 -12.75 -13.74 -12.01 -12.98 Country effect (average of Low - High column) Within-country effect (average of Low - High row) taken into account. We can determine these contributions only to the extent that these sorts do not lead to the same exact division of the BMI sample. Figure 3 plots the fraction of stocks in each of the resulting 25 portfolios in Panel A and the fraction of market capitalization in Panel B. If these sorts led to the same division of the BMI sample, all the firms and the market capitalization would lie on the diagonal. The plots again show that the two effects overlap considerably, especially when measured in terms of market capitalization. For each of the 25 independent doublesorted portfolios, we estimated an ex post beta and CAPM alpha over the full period. In Table 6, the left columns report betas and the right columns report alphas. In the left columns, betas are widely dispersed and tend to increase from top to bottom and from left to right. For example, the first data column reports the betas of portfolios formed from stocks in low-beta countries; reading down the column, portfolio betas within this subsample range from 0.52 to 1.62. Similarly, the fifth data column reports betas of portfolios formed from stocks in high-beta countries, and the range of portfolio betas within this subsample also is large, from 0.65 to 1.67. In contrast, the rows 54\twww.cfapubs.org\b FAJ_March-April_2014.indb 54 -1.94 -0.78 0.54 1.07 -0.50 2.71 3.47 1.08 0.77 0.15 2.7 3.33 3.11 0.74 -0.10 0.74 1.74 1.21 0.06 -1.11 0.06 0.29 0.27 -1.09 -2.41 -1.04 -1.07 0.04 -0.27 -1.40 0.77 1.43 1.45 2.16 2.45 3.43 0.73 0.78 0.88 0.26 -0.42 2.04 -17.95 1.73 display even less variation in portfolio betas than the industry sorts delivered. For example, the \"Low\" row, which consists of portfolios formed from stocks with the lowest betas, shows that betas range from 0.52 (lowest-beta country) to 0.65 (highest-beta country). The statistics at the bottom of Panel A summarize these patterns. The first is a pure country effect, measured as the average of the differences between high-beta and low-beta countries, controlling for stock-level risk. The second is a pure stock effect, measured as the average of the differences between high-beta and low-beta stocks, controlling for country risk. Our ability to find low-risk portfolios through the selection of lowrisk stocks within countries, at a -1.01 reduction in beta, is 50 times as large as our ability to find lowrisk portfolios through the selection of low-risk countries within stock risk groups. These results are consistent with stock-level beta, as estimated from historical data, being a much better predictor of future beta than each stock's country beta is. This time, the incremental predictive power of country-level historical beta is closer to zero, in both economic and statistical terms. 2014 CFA Institute 3/12/2014 9:32:45 PM The Low-Risk Anomaly Figure 3. \u0007 Distribution of Stocks across Portfolios Formed Jointly on Stocks' Betas and Their Country Betas: Developed Markets, November 1989-December 2012 A. Average Distribution of Stocks across Double-Sorted Portfolios Average Fraction of Stocks 0.10 0.05 0 High Low 4 2 3 3 Stock Beta 2 4 High Country Beta Low B. Average Distribution of Market Capitalization across Double-Sorted Portfolios Average Fraction of Market Value 0.10 0.05 0 High Low 4 2 3 3 Stock Beta 2 4 High Low Country Beta Note: Vertical bars denote the time-series average fraction of stocks (Panel A) or fraction of market capitalization (Panel B) in each portfolio. The right panel of Table 1 reports alphas for the double-sorted portfolios. Alphas tend to decrease from top to bottom and from left to right. The average of these differences within each row, the pure country effect, is 6.22%. The average of these differences within each column, the pure stock effect, is 5.40%. Both are statistically significant at the 10% level. These historical data suggest that the low-risk anomaly is present both within and across countries. In short, the same picture that we saw with industries emerges with countries as well. The beta and alpha estimates in Table 6 suggest that pure March/April 2014\twww.cfapubs.org FAJ_March-April_2014.indb 55 55 3/12/2014 9:32:46 PM Financial Analysts Journal stock alpha (the micro effect) and pure country alpha (the macro effect) arise for different reasons. The pure country alpha of 6.22% is present in spite of no difference in cross-country beta. By implication, the country alpha obtains as a consequence of cross-country portfolios exhibiting material differences in returns, with modest differences in risk. In contrast, the pure stock alpha of 5.40% is present alongside a dramatic difference in withincountry beta of -1.01. By implication, the stock alpha obtains as a consequence of within-country portfolios exhibiting material differences in risk, with modest differences in returns. Idiosyncratic Risk The same analysis can be done with idiosyncratic risk instead of beta, but the process requires a bit more decision making. In particular, the industry contribution could be measured with either the idiosyncratic risk of aggregated industry portfolios or the average firm-level idiosyncratic risk within the industry. The former measures the risk of the industry; the latter measures how risky the stocks within the industry are. We opted for the second approach for both practical and theoretical reasons. Practically, our intuition was that the idiosyncratic volatility of the industry average portfolio would perhaps be lower than that of all the individual stocksbecause of the benefits of diversificationand it might be mechanically related to the number of firms within each industry or country. Theoretically, we viewed the demand as arising at the stock level, such that an industry with many high-risk stocks generates more aggregate demand than an industry with few such stocks. The data supported our intuition. We repeated the analysis from Tables 4 and 6 using idiosyncratic risk instead of beta, and we show the summary results in Table 7. The first row of Panel A (Panel B) shows a pure industry effect (pure country effect), measured as the average of the differences between high- and low-idiosyncratic risk industries (countries), controlling for stock-level idiosyncratic risk. The second row of each panel is a pure stock effect, measured as the average of the differences between high- and low-idiosyncratic risk stocks, controlling for industry or country idiosyncratic risk. We found that the idiosyncratic risk effect is almost entirely a micro phenomenon: 96% comes from a pure stock effect in the CRSP industry sample, and 86% comes from a pure stock effect in the BMI country sample. Idiosyncratic risk is similar to valuation ratios in the sense that the intuition behind it does not necessarily hold in the aggregate. What matters more is whether a stock is risky relative to another stock in its category rather than the category average. 56\twww.cfapubs.org\b FAJ_March-April_2014.indb 56 Table 7. \u0007 CAPM Betas and Annualized Alphas for Portfolios Formed Jointly on Prior 60-Month Idiosyncratic Volatility and Prior Industry and Country Idiosyncratic Volatility CAPM Point Estimate t-Statistic A. CRSP sample, 1968-2012 Industry -0.22 -6.31 effect Within-0.73 -14.39 industry effect Total effect B. BMI sample, 1989-2012 Country 0.03 0.43 effect Within-0.78 -12.66 country effect Total effect CAPM (% per year) Point Estimate t-Statistic 0.42 0.22 10.21 3.59 10.63 1.54 0.42 9.72 2.82 11.26 Conclusion: Samuelson's Dictum and the Implications for Managed-Volatility Strategies The decomposition of the low-risk anomaly has both theoretical and practical implications. The superior performance of lower-risk stocks that is evident in Figure 1 in both US and international markets is a very basic form of market inefficiency. It has both micro and macro components. The micro component arises from the selection of lower-risk stocks, holding country and industry risk constant. The macro component arises from the selection of lower-risk industries and countries, holding stocklevel risk constant. In theory, market inefficiency arises from some combination of less than fully rational demand and the limits to arbitrage. Shiller (2001) attributed to Samuelson the hypothesis that macro inefficiencies are quantitatively larger than micro efficiencies. In contrast, we found that the contribution to the lowrisk anomaly is of similar magnitude across micro and macro effects. Whereas stock selection leads to higher alpha through a combination of significant risk reduction and modest return improvements, country selection leads to higher alpha through a combination of significant return improvements and modest risk reduction. Industry selection has a relatively modest effect, and the extra alpha 2014 CFA Institute 3/12/2014 9:32:47 PM The Low-Risk Anomaly cannot be statistically distinguished from zero. We view this finding as an affirmation of a modified version of Samuelson's dictum; namely, micro efficiency holds only up to the limits of arbitrage. The constraint of fixed-benchmark mandates makes low-risk stocks unattractive to many institutional investors. Macro inefficiency is more present for countries than for industries, perhaps because arbitrage is more limited across countries than across industries. The absence of material risk reduction from the macro selection of industries and countries suggests two things. First, compared with predicting the relative risk of individual stocks, pure country or industry risk prediction is hard. Second, to the extent that this anomaly arises because of investor demand for high-risk aggregations of stocks, this demand is in large part backward looking. The pattern of returns suggests that behavioral demand tilts toward sectors and countries that have experienced relatively higher risk in the past. In practice, the incremental value of industry and country selection, even when holding stocklevel risk constant, suggests that the use of a risk model in beta estimation that includes fixed country and industry effects is preferable to simple stock-level sorts on beta or volatility. This has been especially true in global portfolios, where the incremental value of estimating country-level risk is significant. Moreover, because country and industry exposures have generated incremental alpha, a global mandate with somewhat relaxed country and industry constraints would have delivered higher risk-adjusted returns. We thank Ely Spears and Adoito Haroon from Acadian Asset Management LLC for excellent research assistance. Baker gratefully acknowledges the Division of Research of the Harvard Business School for financial support. This article qualifies for 1 CE credit. Notes 1.\thttp://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ index.html. References Ahearne, Alan, William Griever, and Francis Warnock. 2004. \"Information Costs and Home Bias: An Analysis of US Holdings of Foreign Equities.\" Journal of International Economics, vol. 62, no. 2 (March):313-336. Black, Fischer, Michael C. Jensen, and Myron Scholes. 1972. \"The Capital Asset Pricing Model: Some Empirical Tests.\" In Studies in the Theory of Capital Markets. Edited by Michael C. Jensen. New York: Praeger. Ang, Andrew, Robert J. Hodrick, Yuhang Xing, and Xiaoyan Zhang. 2006. \"The Cross-Section of Volatility and Expected Returns.\" Journal of Finance, vol. 61, no. 1 (February):259-299. Blitz, David, Juan Pang, and Pim van Vliet. 2012. \"The Volatility Effect in Emerging Markets.\" Working paper (April). . 2009. \"High Idiosyncratic Volatility and Low Returns: International and Further U.S. Evidence.\" Journal of Financial Economics, vol. 91, no. 1 (January):1-23. Asness, Cliff, Andrea Frazzini, and Lasse H. Pedersen. 2013. \"Low-Risk Investing without Industry Bets.\" Working paper, AQR Capital Management (May). Baker, Malcolm, Brendan Bradley, and Jeffrey Wurgler. 2011. \"Benchmarks as Limits to Arbitrage: Understanding the Low-Volatility Anomaly.\" Financial Analysts Journal, vol. 67, no. 1 (January/February):40-54. Baker, Nardin, and Robert A. Haugen. 2012. \"Low-Risk Stocks Outperform within All Observable Markets of the World.\" Working paper (April). Bali, Turan, Nusret Cakici, and Robert F. Whitelaw. 2011. \"Maxing Out: Stocks as Lotteries and the Cross-Section of Expected Returns.\" Journal of Financial Economics, vol. 99, no. 2 (February):427-446. Basu, S. 1977. \"Investment Performance of Common Stocks in Relation to Their Price-Earnings Ratios: A Test of the Efficient Market.\" Journal of Finance, vol. 32, no. 3 (June):663-682. Black, Fischer. 1972. \"Capital Market Equilibrium with Restricted Borrowing.\" Journal of Business, vol. 45, no. 3 (July):444-455. Brennan, Michael. 1993. \"Agency and Asset Pricing.\" Working paper, University of California, Los Angeles (May). Campbell, John. 1991. \"A Variance Decomposition for Stock Returns.\" Economic Journal, vol. 101, no. 405 (March):157-179. Daniel, Kent, and Tobias Moskowitz. 2013. \"Momentum Crashes.\" Working paper, Columbia University (April). Falkenstein, Eric. 1994. \"Mutual Funds, Idiosyncratic Variance, and Asset Returns.\" Dissertation, Northwestern University. Fama, Eugene F., and Kenneth R. French. 1992. \"The Cross-Section of Expected Stock Returns.\" Journal of Finance, vol. 47, no. 2 (June):427-465. . 1997. \"Industry Costs of Equity.\" Journal of Financial Economics, vol. 43, no. 2 (February):153-193. . 1998. \"Value versus Growth: The International Evidence.\" Journal of Finance, vol. 53, no. 6 (December):1975-1999. Frazzini, Andrea, and Lasse H. Pedersen. 2010. \"Betting against Beta.\" Working paper, New York University (May). French, Kenneth R., and James M. Poterba. 1991. \"Investor Diversification and International Equity Markets.\" American Economic Review, vol. 81, no. 2 (May):222-226. March/April 2014\twww.cfapubs.org FAJ_March-April_2014.indb 57 57 3/12/2014 9:32:47 PM Financial Analysts Journal Haugen, Robert A., and A. James Heins. 1975. \"Risk and the Rate of Return on Financial Assets: Some Old Wine in New Bottles.\" Journal of Financial and Quantitative Analysis, vol. 10, no. 5 (December):775-784. Hong, Harrison, and David Sraer. 2012. \"Speculative Betas.\" Working paper, Princeton University (August). Jung, Jeeman, and Robert J. Shiller. 2005. \"Samuelson's Dictum and the Stock Market.\" Economic Inquiry, vol. 43, no. 2 (April):221-228. Karceski, Jason. 2002. \"Returns-Chasing Behavior, Mutual Funds, and Beta's Death.\" Journal of Financial and Quantitative Analysis, vol. 37, no. 4 (December):559-594. Kothari, S.P., and Jay Shanken. 1997. \"Book-to-Market, Dividend Yield, and Expected Market Returns: A Time-Series Analysis.\" Journal of Financial Economics, vol. 44, no. 2 (May):169-203. Lamont, Owen, and Jeremy C. Stein. 2006. \"Investor Sentiment and Corporate Finance: Micro and Macro.\" American Economic Review, vol. 96, no. 2 (May):147-151. 58\twww.cfapubs.org\b FAJ_March-April_2014.indb 58 Lewellen, Jonathan. 1999. \"The Time-Series Relations among Expected Return, Risk, and Book-to-Market.\" Journal of Financial Economics, vol. 54, no. 1 (October):5-43. Li, Xi, Rodney N. Sullivan, and Luis Garcia-Feijo. 2014. \"The Limits to Arbitrage and the Low-Volatility Anomaly.\" Financial Analysts Journal, vol. 70, no. 1 (January/February):52-63. Samuelson, Paul. 1998. \"Summing Up on Business Cycles: Opening Address.\" In Beyond Shocks: What Causes Business Cycles? Edited by Jeffrey C. Fuhrer and Scott Schuh. Boston: Federal Reserve Bank of Boston. Scholes, Myron. 1972. \"The Market for Securities: Substitution versus Price Pressure and the Effects of Information on Share Prices.\" Journal of Business, vol. 45, no. 2 (April):179-211. Shiller, Robert J. 2001. Irrational Exuberance. New York: Broadway Books. Vuolteenaho, Tuomo. 2002. \"What Drives Firm-Level Stock Returns?\" Journal of Finance, vol. 57, no. 1 (February):233-264. 2014 CFA Institute 3/12/2014 9:32:47 PM Financial Analysts Journal Volume 70 Number 3 2014 CFA Institute PERSPECTIVES High-Frequency Trading and Its Impact on Markets Maureen O'Hara At the 2013 CFA Institute Financial Analysts Seminar, held in Chicago on 22-25 July, Maureen O'Hara discussed a new market paradigm: Trading has become faster, and market structure has fundamentally changed. In today's market, high-frequ