Dynamic Risk Management: Balancing Target-Date Risks for Changing Market Environments

Transcription

1 INVESTMENT VIEWPOINTS JUNE 2017 Dynamic Risk Management: Balancing Target-Date Risks for Changing Market Environments Executive Summary Evolving our approach to managing risk along the glide path. Downside protection and risk management have always been hallmarks of our approach to target-date management. Now we are evolving our glide path to better balance risks for participants relative to the prevailing market regime, beyond just their age. Our overall philosophy and framework remain intact, and our analysis shows that the changes have an across-the-board positive effect on overall performance. Radu C. Gabudean, Ph.D. Vice President Portfolio Manager Multi-Asset Strategies Nancy Pilotte, CAIA Vice President Client Portfolio Manager Multi-Asset Strategies Dynamic risk management, as opposed to tactical allocation. We evaluate the market environment using a medium-horizon, multi-factor signal. We shy away from a short-term signal, or one focused solely on maximizing total return. Our model attempts to capture cyclical turns, creating a desirable mean-reverting behavior. Asymmetric, not naïve, changes along the glide path. In contrast to naïve approaches that make parallel adjustments all along the glide path regardless of age, we employ age-specific changes around our existing strategic glide path. Our adjustments are asymmetric, owing to the uneven distribution of both wealth and risk along the glide path. Specifically, we allow for more equity exposure early in an investor s life, while constraining the equity allocation in the late-career and in-retirement periods; equity allocations can rise or fall for investors in mid-career. Retaining comparatively flat glide path shape. An assessment of the relative performance of stocks versus bonds allows us to make informed judgments about the likely path of future asset returns. This enables us to essentially trade sequenceof-returns risk for improvements in other risk metrics when and where it makes sense to do so along the glide path. Nevertheless, our beneficial flatter glide path shape remains intact for the critical years around retirement, where it has been among the flattest in the industry. Latest enhancement designed to increase likelihood of success for greatest number of participants. Our research shows that dynamic risk management may provide higher wealth at retirement, increase the probability of retirement success, and reduce risk relative to the static, strategic glide path. We believe this approach represents a distinct improvement not only over static glide path designs but also over competing, more tactical, active designs. Rich Weiss Chief Investment Officer Multi-Asset Strategies

2 Introduction Economies cycle. Policies shift. Markets ebb and flow. Can we improve our target-date portfolios by being more responsive to market conditions? Our objective has always been to increase the likelihood of retirement success for the broadest number of participants. We ve followed that objective by balancing the multiple risks that retirement investors face over time. To improve that balance, we systematically evaluate our asset allocation, sub-asset classes, and manager selection. Consistent with this method, we are incorporating a dynamic approach to risk management that follows the market environment. In this paper, we will explain the rationale behind this evolution in our approach, show the research supporting this conclusion, and lay out the intended benefits for target-date fund (TDF) investors. We ve broadened our existing framework for managing TDF risk by incorporating information about the market environment over a medium-term horizon. Where We ve Been Prior to this enhancement, we employed a well-defined, long-term strategic glide path in our target-date portfolios that incorporated long-term market assumptions of risk and return. We have produced a large body of literature detailing the extensive proprietary work and practitioner research involved in the construction of this glide path. This work is documented most recently in Pilotte and Weiss, We proceed from the view that a glide path must be robust enough to account for a broad range of withdrawal assumptions by providing enough equity exposure to limit longevity risk. On the other hand, we believe glide path slope and equity allocations around retirement also must limit market and tail risks for those at or near retirement when account balances are typically highest. Indeed, this approach of a comparatively flat glide path before retirement and fully flat path in retirement has generated attractive risk-adjusted performance over time....and Where We re Going As we outline in this paper, our latest research into improving the success rate for TDF investors provides us with another tool to achieve a better balance of risks for investors. We adjust the balance based on the economic and market environment because each environment carries with it different payoffs to different risks, just as each age carries with it different sensitivities to different risks. As a result of this analysis, we have elected to broaden our existing framework for managing risk by incorporating information about the market environment over a medium-term horizon into our glide path design. Our solution represents a step forward from having allocations expressed primarily as a function of age (where age is a proxy for investor wealth, time horizon, and risk tolerance, among other factors) and uses a combination of age and environment. Importantly, the environment modifies allocations differently over the lifecycle due to its uneven effect on participants of various ages. We demonstrate that the investment policy of a TDF does well to respond to the environment and, importantly, the response should treat the entire glide path as one pool of wealth moving through time and not as a disparate set of portfolios. Thus, the response becomes differentiated by a participant s age and varying sensitivities to lifecycle investing risk over time. Using various fundamental, technical, and economic variables, we categorize different market environments based on their relative favorability to stocks versus bonds. Our prior long-term, strategic glide path remains central to our approach, forming our neutral position. For each market environment defined, we determine the stock/bond allocation adjustments for each year on the glide path, taking into account the interaction of market environment with other sources of risk for each age cohort. Ultimately, our solution specifies seven related glide paths three reflecting a positive equity environment and three a negative equity environment either side of our neutral or long-term strategic glide path. 2

3 Part I: Lifecycle Risks Change With the Market Environment The balance of risks in lifecycle investing is a function of an investor s age and the market environment. We define the market environment as the degree of relative attractiveness of stocks versus bonds, and for a given market environment, we tilt the glide path to better balance these risks. We illustrate the multiple risks inherent in lifecycle investing in Figure 1 below and discuss them in detail in Pilotte and Weiss, Figure 1: Understanding the Balance-of-Risk Framework Longevity Risk: Risk of running out of money in retirement Market Risk: Dispersion in outcomes caused by changes in the level of market prices Inflation and Interest Rate Risk: Dispersion in outcomes due to rising inflation and/or interest rates Sequence-of-Returns Risk: Dispersion in outcomes caused by the sequence, rather than the level, of market returns Source: American Century Investments. Tail Risk: Dispersion in outcomes caused by sudden, extreme changes in market prices Abandonment Risk: Dispersion in outcomes caused by poor timing of exit from an investment strategy Longevity risk, or the risk of running out of money in retirement, is paramount. But longevity risk must be balanced against other risks, such as market risk and tail event risk, which can have adverse effects on account balances, or abandonment risk, in which participants may make poorly timed sell decisions or abandon a saving and investing plan altogether. These risks are influenced, and some even defined, by changing market conditions over time. Thus, it makes intuitive sense that taking this dynamic environment into account should improve the balance among these various risks. We guide the search for a solution with our balance-of-risk framework. This method ensures a holistic result that treats the glide path as one entity, instead of looking at each vintage portfolio separately. As a further benefit, we naturally use success metrics relevant and specific to TDF investors, instead of metrics applicable to more generic portfolios such as one-period returns or Sharpe ratios. 3

4 Our solution requires two components: a glide path for each environment and a model to identify the environment. The two must work together. Forecasting the Market Environment Our solution to a shifting market environment requires two parts: a glide path, or set of allocations to stocks and bonds across ages for each environment, and a model to identify the environment. The two cannot be thought of in isolation because they must be designed in tandem to serve our ultimate purpose: improving the long-term retirement success for the greatest number of participants. We begin our analysis by evaluating the relative attractiveness of stocks versus bonds from the perspective of longer-term trends using a proprietary multi-factor model. This model is built on economically sensible factors relating to macroeconomic conditions, valuation, and technical conditions. It forecasts the return of stocks versus bonds over a three-year horizon. When designing the model, we balanced the need to cover the important determinants of the relation between stocks and bonds with the pitfalls of a complex model. Figure 2 displays the fit of this model to realized values over the next three-year period. As we can see, the forecast captures the major turns in realized returns. Figure 2: Modeled Forecast Versus Realized Values: Excess Returns of Stocks Over Bonds Modeled Forecast of 3-Year Excess Returns Following 3-Year Realized Excess Returns Log Return Difference, Annualized 30% 20% 10% 0% -10% -20% The model aims to capture the upcoming cyclical turns, creating a mean-reverting behavior. The special nature of the TDF investor allows for this beneficial longer- term focus. 1 See Cochrane, 2008 for a review of the relationship between time horizon and return predictability. -30% Source: American Century Investments. FactSet, FRED. Data as of 3/31/2017. This hypothetical situation contains assumptions that are intended for illustrative purposes only and are not representative of the performance of any security. There is no assurance similar results can be achieved, and this information should not be relied upon as a specific recommendation to buy or sell securities The medium-term horizon of the model captures the upcoming turns of a market or business cycle, which leads to its mean-reverting behavior. It dislikes equities after a long bull market, or when relative valuations to bonds become poor by historical standards, or when it identifies an economic peak. We chose this medium-term focus as academic studies show that accuracy of return predictability increases with the horizon. 1 The long investment horizon of a TDF investor allows for such a desirable choice, in contrast with more generic allocation funds that may be judged more on their short-term performance. As an added benefit, this mean-reversion feature balances investors natural tendency for momentum chasing. They may load up on risk at the market peak and, conversely, be most fearful at market bottoms, as evidenced by numerous academic and practitioner studies of investor behavior. Following the tenets of long-term investing, our model guides them in the opposite direction, pulling equity exposure down when we believe markets are frothy and adding exposure after a sell-off in equities

5 Our signal changes slowly over time, and our three-tier structure limits any extreme readings of the model. We use this forecast together with information about the level of risk to create a definition or signal for the market environment, which indicates both direction (positive or negative on stocks relative to bonds) and degree of conviction (relative attractiveness). This signal can express three degrees of conviction on either side of neutral, ranging from very unfavorable to very favorable for stocks. Figure 3 shows that our signal changes slowly over time, reflecting the comparatively long-horizon nature of our forecast. For example, the most recent signal peaks have been in 2014 and 2004 respectively, a time span of a decade. Finally, our translation into the threetier setup provides a structure that limits any extreme readings of the model, even when factors take on extreme values. 3 2 Figure 3: Environment Definitions Based on Modeled Excess Returns Forecast and Volatility Equity Signal Tier Source: American Century Investments. Data as of 3/31/2017. This hypothetical situation contains assumptions that are intended for illustrative purposes only and are not representative of the performance of any security. There is no assurance similar results can be achieved, and this information should not be relied upon as a specific recommendation to buy or sell securities. Part II: From Market Environment Assessment to Glide Path Implementation 2010 Each signal and tier regime displayed in Figure 3 is associated with a degree of excess return and risk expectations for stocks relative to bonds. For each of those environments, we assign an associated glide path. We analyze the glide path for each environment relative to the neutral glide path position, in order to provide better intuition about the relation among these environments A neutral score of 0 is associated with our long-standing strategic glide path. Positive signals connote environments with above-average performance expectations for equities relative to bonds over an intermediate horizon. As a result, the associated glide paths will have greater equity exposure than our long-term strategic allocation. Negative scores, indicating less attractive performance for equities going forward, would have less equity exposure. An Asymmetric Approach After reviewing several options as part of our analysis, we settled on allocations that reflect market conditions but also better express investor preferences along the glide path that is, glide paths that are designed to offer greater downside protection near retirement; greater acceptance of market risk in the years far from retirement; and a more nuanced approach mid-career. The resulting allocations are illustrated by the range of glide paths depicted in Figure 4, representing the seven tiers or market environments. 5

6 Figure 4: Our Dynamic Risk Management Approach Adjusts to a Variety of Market Environments Tier +3 Tier +2 Tier +1 Strategic (Anchor) Glide Path Tier -1 Tier -2 Tier -3 Blanket, simplistic solutions modifying equity exposures without regard to age, account balance, and risk tolerance are misguided. To achieve a better balance of risks, an asymmetric approach is necessary. Equity Allocation 100% 80% 60% 40% 20% 0% Early Career Mid Career Late Career In Retirement Grounded by Strategic Glide Path Increase equity only before age 60 Decrease equity only after age Age Source: American Century Investments. In our view, blanket solutions that uniformly add or subtract equity exposures without regard to age, account balance, and risk tolerance are misguided. Even more complex versions where the equity range is scaled by age (e.g., older investors experience half the change of the younger ones) do not fully reflect the changing composition of risks along the glide path. As part of this study, we investigated such approaches and found them wanting. To achieve a better balance of risks, we believe an asymmetric approach is necessary. Specifically: For environments favorable to stocks relative to bonds, young participants stay allocated at maximum equity for several years longer than the anchor, or strategic, glide path would indicate. Plan participants in mid-career also maintain a higher allocation to equities that gradually converges toward the anchor, while investors close to or in retirement keep the same allocation as the anchor. For environments unfavorable to stocks relative to bonds, young investors stay with the anchor allocation. Mid-career investors de-risk increasingly with age, such that they may reach the anchor s retirement allocation several years before retirement. Participants approaching or already in retirement see their equity allocation drop below that of the strategic glide path. We do not go beyond the stock-versus-bond decision and express changes across all underlying sub-asset class allocations in proportion to their weighting within the asset class. The rationale for this proportional approach is that the broad asset allocation is the principal determinant of risk, especially at the horizon we investigate here. We view dynamic allocations within an asset class, e.g., value versus growth equities, to be more the purview of tactical asset allocation, something we do not do in our target-date portfolios (For a more detailed discussion of our view on sub-asset class allocation along the glide path, see Pilotte and Weiss, 2016.) To summarize, our asymmetric solution addresses competing risks along the glide path, expressed in several features that we deem important for risk management, which we will view and present next through multiple lenses. Above, we describe the allocations by our model s signal tiers. In the next section, we discuss extensively our allocations across age groups and touch on the contrast between dynamic risk management and tactical allocation. 6

7 Our dynamic risk management approach differs from a typical tactical allocation through our longerterm signal, asymmetric implementation, and adjustments only at the broad stock/bond level. Early-career participants must stay at high equity allocation to invest their important new contributions, thus addressing longevity risk. Why a Dynamic Risk Management Approach Is Preferable to Tactical Allocation We see three primary differences between our dynamic risk management approach and typical tactical allocation. First, our model is a longer-term, slow-moving signal designed to capture higher-level trends and mean-reverting behavior by stocks and bonds in relation to the overall business and economic cycle. Second, our implementation is asymmetric. That is, we take into account age, wealth, risk tolerance, and time horizon for investors at each point along the glide path. Third, our glide path adjustments are at the overall stock/bond level, with no changes to the proportions across underlying sub-asset classes. In stark contrast, tactical allocation approaches typically emphasize short-horizon opportunities at the sub-asset class level, particularly in emerging markets, small-caps, and commodity-related strategies. Such tactical decisions typically focus solely on maximizing total return and essentially may double down on short-term beta trends and easily overlook higher-level trends in the broader market environment. Moreover, tactical allocation overlays are naively bolted uniformly across the glide path without regard to the specific risk tolerances. Keeping to a medium-term, multifactor, mean-reverting approach to broad stock/bond allocation decisions anchors our approach firmly in the realm of risk management. An Asymmetric, Dynamic Approach: Adjusting the Glide Path by Age The key concerns and risks for investors differ according to their age, account balance, and associated factors. Such asymmetry of risk along the glide path necessarily demands an asymmetric solution. For example, younger investors are more sensitive to longevity risk, but the higher equity allocation that best addresses this concern is inappropriate for investors at or near retirement, when market risk predominates. Said differently, higher equity allocations have the effect of increasing the chance of success for the average participant (the median account balance at retirement is higher). But at the same time, the spread, or dispersion, of outcomes around that median is also greater there is a much bigger gap between winners and losers. In addition, investors are much more sensitive to a loss than they are to an equivalent gain, particularly in the crucial years around retirement, when account balances are typically highest. This means that the effect of a market downturn will also be felt asymmetrically. As a consequence of all these elements, young participants risk sensitivities and preferences are in many ways counter to those of investors at retirement. Early-career investors. For participants ages 35 and under, we take advantage of environments favorable to equities to increase the equity exposure up to the strategic glide path s maximum of 85%, staying at that maximum for longer. However, we do not act upon the negative equity signal. This is because the impact of ongoing contributions and regular rebalancing outweighs any benefits from downside protection for the youngest participants, for whom longevity risk predominates. Our approach for young investors seems appropriate when we consider the nature of our signal and potential anticipated or unanticipated poor market returns. The meanreverting feature ensures that we are increasing our equity exposure after a period of poor equity returns, or at market bottoms, when valuations seem attractive. Should our signal be incorrect for a given period and poor equity performance continue for longer than anticipated, this will simply provide a better investment opportunity for their fresh contributions. These new contributions make a large impact on their still-small portfolio, and they should be invested for the high return provided by an equity-heavy portfolio. A similar analysis leads us not to take advantage of the negative-equity signal and not to regret when the signal turns out to be correct and equities perform poorly. 7

8 2 We analyzed the last 20 and 40 years of data, using both nominal and excessof-cash returns. Note that the 85% of the risk of an all-equity portfolio result is not simply because we have an 85% allocation to equities. It is a result of the correlation and volatility levels of stocks and bonds over time. Mid- to late-career participants can take full advantage of market signals benefiting from downside protection provided by a negative signal and the potential for higher return provided by a positive one. Near- and in-retirement participants use the signal solely to protect their existing wealth, avoiding any increases in equity risk. We choose to limit the equity exposure in any environment at the maximum of 85% provided by the strategic glide path. At 85% equity exposure, an equity/bond portfolio has historically delivered 2 around 95% of the returns of an all-equity portfolio with only 85% of the risk. This higher risk/return trade-off marginalizes the benefit of higher equity exposures. Intuitively, this result comes from several factors, all important: rebalancing benefits, better risk-adjusted performance by bonds, and plain-old diversification. For investors 25 and under, we stay at the neutral equity exposure of 85% there is no further adjustment. For investors between the ages of 25 and 35 we may delay the start of reduction in equity exposure from the maximum of 85% versus the neutral allocation start. Reductions in equity allocation below the neutral levels can occur after age 35 as market risk looms larger, but initially equity increases are larger in magnitude than decreases while longevity risk continues to dominate. Mid- to late-career investors. For investors ages 36 to 60, the dynamic adjustments can vary up or down from the strategic glide path weights, taking full advantage of market signals. In good market environments, these participants will reduce their longevity risk through higher returns, while in poor market environments, a reduction in market risk will buffer potential losses. The maximum overweight potential in favorable markets for equities is roughly 8% above the strategic glide path weighting, occurring at age 37 for a maximum weight of 85%. The typical overweight is much smaller than that, at around 2% for a moderate signal. The potential to underweight equity increases, with the maximum underweight reaching 9% at age 56, which would result in a 45% equity allocation. The typical underweight is much smaller than that, at around 2.5% for a moderate signal. Investors near or in retirement. For investors ages 60 and beyond, when account balances tend to be large and aversion to loss is high, our approach is skewed toward downside protection. Older investors with little time remaining before they must begin withdrawing their savings necessarily place more focus on preserving their wealth by avoiding equities possibly large negative returns. Therefore, we allow the allocation to go below the 45% anchor allocation because we want participants at retirement to take advantage of the negative equity signal at the precise moment when it makes the largest absolute impact for them. The downside-only focus feels even more appropriate when we consider the meanreverting nature of our signal: We will only reduce retirees equity exposure, and that will likely occur after an equity market run-up or after a good economic stretch, when their balances are likely to be high. Should our signal be wrong and the market rally continue, we are only giving up a modest amount of upside because equity exposure remains rather significant even at the minimum allocation (40%). No overweight beyond the strategic equity target is allowed after age 60 to maintain the focus on downside protection, while the potential to underweight equity increases. For participants in retirement (65 or older), equity allocation can vary between the strategic glide path weight of 45% and a minimum allocation of 40%. 8

9 We use these judgments about the likely future course of returns to tilt the balance of risks further in investors favor precisely when it makes most sense to do so, trading sequence-of-returns risk to benefit market and tail risks. Part III: The Effects of Asymmetry on Glide Path Slope and Sequenceof-Returns Risk We have long maintained that a flatter to retirement glide path is preferable to a steeper path running through retirement, all else equal. Here we should acknowledge that our risk management process may intuitively have the effect of steepening the slope of the glide path. This is primarily true in the early and mid-career sections of the glide path for the most extreme market environments. However, the flatter shape remains intact for the critical years around retirement. Indeed, our long-term strategic glide path remains among the flattest in the industry even after allowing for these asymmetric adjustments, and we continue to reach our most conservative equity allocation at retirement. This is demonstrated in Figure 5, which shows that the slope of our glide path retains on average substantially the same flat shape. 95% Figure 5: Our Target-Date Glide Path Remains Among the Flattest in the Industry 85% Strategic Glide Path Equity Dynamic Glide Path Average Equity Industry Average Equity Equity Allocation 75% 65% 55% 45% 35% 25% Source: American Century Investments. Morningstar, Inc Years to Retirement Rather than focus on glide path slope in isolation, however, we encourage readers to consider how incorporating the market environment into the risk calculus can provide a better balance of risks. Our judgments about the market environment over the intermediate-term horizon are implicit statements about the likely sequence of returns in future years. We are in effect using these judgements about the anticipated path of returns to tilt the balance of risks further in investors favor precisely when it makes most sense to do so given their age, account balance, and investment horizon. For example, if our model suggests a more treacherous path ahead for equities, we can reduce market risk for investors who benefit most those near or in retirement. In this example, we have intentionally steepened the glide path (traded sequence-of-return risk) to benefit market and tail risk. Later, we can readjust the glide path when the outlook for equities has improved. Let s look at the problem a different way our own analysis and a large body of thirdparty research demonstrate that flatter glide paths are preferable to steep ones, and that to glide paths are preferable to through retirement glide paths, all else equal i.e., in retirement a flat allocation is best. These outcomes are largely a result of the fact that to TDFs with their flatter glide paths tend to take less market risk and sequence-of-returns risk in the crucial years around retirement, and produce less volatility, meaning greater certainty around participant retirement outcomes. These are precisely the effects we seek to accentuate when equity risk is elevated, and none of the changes we make violate these principles. Our equity allocation continues to

10 TDFs with steep glide paths and no market view are accentuating sequence-of-returns risk for no benefit in terms of longevity risk or market risk. 3 For a framework on how to build such measures, see Gabudean, 2015, which focuses more on characteristics rather than directly on risk. For a more riskoriented example, see Weiss and Wittman, Measuring retirement outcomes is complicated by a lack of consensus on a well-defined set of measures. We propose a dashboard of metrics to capture the experience before, at, and in retirement. 4 For more details and an illustration see Figure 9 in the Appendix. reach its most conservative weighting at retirement and cannot add risk relative to our strategic glide path thereafter. Finally, consider this argument: TDFs with steep glide paths and no market view are accentuating sequence-of-returns risk for no benefit in terms of longevity risk or market risk for their investors. Our own detailed discussion of glide path slope and sequence-of-returns risk can be found in Pilotte and Weiss, Part IV: A Dashboard for Defining Retirement Success Our effort aims to improve retirement outcomes for the broadest number of participants. To judge the success of our proposed solution, we need to measure these outcomes. This task is complicated by a lack of consensus on a well-defined set of measures. 3 In this section, we will describe a set of metrics that are relevant to the retirement risks shown in Figure 1 and have broad applications for measuring participant success beyond our own research shown here. Later, we track their changes to show how our dynamic risk management approach stands to improve outcomes. Notably, we use real that is, after-inflation returns throughout our analysis to better reflect changes in outcomes over time, net of inflation s two-sided effects. As a result, we do not segregate inflation as a separate risk. Lastly, these measures are influenced by various assumptions about capital markets, contributions, and withdrawals. Our Dashboard Risk Measures The risk measures on our dashboard can be grouped into two broad categories: Measures of wealth distribution. We measure wealth at various points along the glide path, with emphasis on metrics both at and in retirement. Such measures include: average, median, standard deviation, and various percentiles of wealth at a given time. They can be further combined into, for example, a risk-adjusted performance measure, as shown in the Appendix. These measures indicate the status of plan participants at various ages, like a set of pictures taken at various points in a race. Measures of the participant experience up to retirement. The path taken by the investor s portfolio can be measured by analyzing the maximum drawdown or worst returns and various scenarios. Beyond the point-in-time status shown by the previous set of measures, we want to gauge investors ongoing experience. While there is a relationship between wealth at various points and the investor experience preceding it, we may be missing some issues by using just the former measure. This is an important, if perhaps easily overlooked, point many roads may lead to the same destination, but they may not be equally pleasant to travel. Wealth distribution alone may not capture any issues related to abandonment or simple anxiety on the investor s part for enduring a rough ride. Measuring Wealth at Retirement Wealth distribution measures at retirement 4 capture many important risks for targetdate investors: Longevity. The higher the wealth at retirement, the greater the chances to successfully retire. Further, the higher the wealth at retirement in worst-case scenarios, the higher the number of investors who will retire successfully. Sequence of returns. Investment strategies more exposed to this risk will result in a greater dispersion of wealth outcomes at retirement. Tail events. Investors who experience an unusually adverse market will see their wealth at retirement register at a very low level. As a result, the low percentiles of the wealth distribution should reflect the potential effect of such tail events, particularly when they occur close to retirement. 10

11 One measure may speak at the same time to several risks. We do not attempt to disentangle the various risks within our metrics because these metrics are better suited to gauge what ultimately matters: retirement success. 5 See Pfau, 2016, for a discussion. 6 Abandonment can happen by the plan participant or by the actual plan! How likely are boards to keep underperforming strategies in the plan? There is a robust history of research documenting ill-timed investor sell decisions, up to and including the most sophisticated institutional investors (see Goyal and Wahal, 2008). Abandonment among TDF investors has also been identified but occurs at a much lower rate than equity-only investors (see Ameriks, Marshall, and Ren, 2009). Lei and Yao, 2015, identified wealthier, loss-averse investors as more likely to make market timing mistakes during down markets. Intuitively, market losses are felt most keenly around retirement, when risk aversion and account balances are highest, and investors switch from contributions to withdrawals. Equity market and interest rate risk. Environments where equities or bonds have returns persistently below long-term averages decrease wealth at retirement and can increase the dispersion of wealth outcomes. We see that a particular measure may speak to several risks at the same time. For example, low wealth percentiles may reflect a sudden tail event or a poorly ordered sequence of returns. We do not attempt to fully disentangle the various risks within our metrics. What ultimately matters is a successful retirement, and these metrics are better suited to measure that success over an entire population. Ultimately, we ensure that we use enough metrics to cover the entire universe of risks as we understand and define them today. Measuring Success in Retirement Other wealth distribution measures describe the experience in retirement. Besides the various wealth percentiles, we show a measure derived from the distribution of wealth, namely probability of success. The likelihood that wealth is positive at a certain age is the same as the likelihood of a successful retirement up to that age at least in a simplified world where we don t worry about leaving money for the next generation. The metrics about wealth in retirement depend heavily on withdrawal assumptions, an issue that does not afflict at-retirement results. The various withdrawal assumptions affect these metrics more than any reasonable change in any other assumption, such as the return and risk of various assets, or the contributions. Researchers have used various assumptions, 5 from simple ones to more involved setups believed to be closer to reality. However, more complex assumptions come with more complex behavior, which require more complex metrics to track and compare. To avoid overly complex setups, we proceed from two simple assumptions: (1) withdrawals as a fraction of average of wealth at retirement (W 65 ); and (2) a fraction of the value of W 65 for that scenario. Reality may prove to be somewhere in between these two. Values resulting from these assumptions are shown in the Appendix, as Figures 10 and 11. Measuring the Participant Experience Before Retirement The final set of measures captures the experience before retirement. When we compute the wealth distribution measures, we conveniently assume that a participant will faithfully contribute a set amount through his working life. Yet, that assumption may prove fanciful in the face of a large drop in portfolio value. Participants may stop contributing or abandon the product altogether if the return experience becomes too uncertain. 6 To measure the pre-retirement experience, we look at: Expected worst drawdowns. Drawdowns can be viewed in either relative terms or absolute dollar values relative to the previous high account balance. The former are typically more pronounced early in the investment lifecycle, when allocations are equityheavy and cash balances are comparatively small. Declines in account values absolute dollar amount are of increasing importance in the years around retirement, when wealth is highest. The higher these drawdowns, the more likely participants are to abandon their saving and investing plan by moving to cash. Scenario analysis. Here, we attempt to capture the effect of market risk, tail risk, and sequence-of-returns risk on the investor experience over time. For example, we look at what happens to investor outcomes under different stock and bond return regimes, such as when a drop in equity returns is followed by a full recovery over a certain number of years. These scenario analyses are dependent upon the age at which the return shock occurs. The path of market returns on the way to retirement matters greatly, with the absolute level of equity exposure and the slope of the glide path being crucially important considerations. The slope is important because in scenarios where the glide path de-risks too quickly, we may miss much of the recovery. 11

12 We must ultimately show that this risk management solution does reduce the various types of TDF risks. We achieve that with our metrics dashboard. 7 95% of scenarios provide a wealth level at retirement higher than this one. Put it differently, it is the best of the worst 5% of outcomes. Part V: Results of Our Analysis Show Dynamic Risk Management Improves Outcomes Across Our Retirement Dashboard We have already described the intuition behind our dynamic risk management solution, discussing its two parts: the environment signal and associated glide paths. However, we must ultimately show that this solution, with its two parts working together, does reduce the various types of risks faced by a plan participant. To that end, we compute the various metrics on our retirement success dashboard and compare their values for our dynamic solution versus our long-term strategic glide path. We should note that for a performance analysis of a model, researchers usually employ a historical back-test analysis. In our case, with each observation requiring up to 70 years of data, such analysis is impossible. Instead, we simulated the annual behavior of our model with a set of assumptions detailed in the Appendix, and we investigated all metrics described above. Beyond the mentioned assumptions, we conducted robustness tests with different sets of assumptions. We show and discuss below a representative sample of our most important metrics. Figure 6 presents the changes experienced in the wealth distribution when we compare dynamic risk management with a static glide path. Significantly, we see a modest improvement across all the at-retirement wealth metrics. These improvements are comparatively modest in percentage terms, but even the small increase in wealth for the 95th percentile 7 of wealth scenarios at retirement amounts to nearly an extra year of contributions. Relative Change From Static Glide Path 4% 3% 2% 1% Figure 6: Dynamic Risk Management May Improve Wealth Measures at Retirement 0% Median Wealth 90th Percentile 95th Percentile 99th Percentile Sharpe Ratio of Wealth Source: American Century Investments. This hypothetical situation contains assumptions that are intended for illustrative purposes only and are not representative of the performance of any security. There is no assurance similar results can be achieved, and this information should not be relied upon as a specific recommendation to buy or sell securities. In the average scenario, wealth improves only slightly (median wealth, Figure 6). However, the dispersion around that average scenario is smaller (dispersion of wealth, Figure 7). Said differently, our research shows that dynamic risk management generates higher wealth and a greater certainty around outcomes relative to a static glide path. Furthermore, Figure 7 shows that participant experience is improved by smaller expected maximum drawdowns both in percentage and particularly in absolute terms. Again, large absolute drawdowns are important because they are typically experienced by participants close to retirement, when account balances, market risk, tail risk, and abandonment risk all tend to be greatest. As a result, this finding is particularly 12

13 beneficial. Note, too, the meaningful increase in risk-adjusted change in wealth, (Sharpe Ratio of Wealth in Figure 6) a concept similar to the more familiar Sharpe ratio of returns. Dynamic risk management may generate higher wealth and a greater certainty around outcomes relative to a static glide path... Relative Change From Static Glide Path 0.0% -1.5% -3.0% -4.5% -6.0% Dispersion of Wealth Figure 7: Dynamic Risk Management May Improve Critical Risk Measures Maximum Drawdown % Maximum Drawdown $ Source: American Century Investments. This hypothetical situation contains assumptions that are intended for illustrative purposes only and are not representative of the performance of any security. There is no assurance similar results can be achieved, and this information should not be relied upon as a specific recommendation to buy or sell securities. Next, we look at the experience in retirement. For simplicity, we assume a withdrawal rate of 4% of the average of wealth at retirement (W 65 ). Hence, the withdrawal amount does not vary across investors or over time. Here, too, we find that a dynamic risk management strategy increases the probability of success in retirement, as shown in Figure 8, particularly at ages 85 and 95. At those ages, success chances are notably smaller than 100%; therefore, our dynamic solution is more likely to show a difference....ultimately delivering the potential for higher chances of retirement success. Difference From Static Glide Path 1.5% 1.0% 0.5% Figure 8: In-Retirement Metrics Show Greater Probability of Success 0.0% Probability of Success by Age Source: American Century Investments. This hypothetical situation contains assumptions that are intended for illustrative purposes only and are not representative of the performance of any security. There is no assurance similar results can be achieved, and this information should not be relied upon as a specific recommendation to buy or sell securities

14 The effect on in-retirement metrics results from dynamic risk management being applied throughout the participant s lifetime, both before and in retirement. The improvements in retirement success come from a better balance of risks throughout the glide path, even though the actions taken around retirement, when wealth is highest, have the largest effect. The results discussed in this section make it clear that dynamic risk management benefits all participants, especially those around retirement. We presented a solution to adjust our allocation to the market environment. Its various elements, many of them novel, produce smoother and more successful payoff streams for the greatest number of participants. Conclusion Lifecycle investing necessarily requires paying attention to multiple risks that affect the objective of providing a successful retirement for plan participants. We believe that our existing glide path provides an appropriate balance among these risks to meet that objective. Nevertheless, we systematically and continuously evaluate our investment processes looking for opportunities to evolve our approach in meaningful ways. In this paper, we detailed how we specifically evaluated how to incorporate market information into the glide path, and we believe that we have found a suitable way to do so an asymmetric approach that treats the glide path holistically. It does so by recognizing that participant risks are unevenly distributed across the glide path and unevenly distributed across different market environments. More precisely, for environments favorable to stocks, we increase the equity allocation of early and mid-career participants, up to a certain maximum, while leaving allocations unchanged for participants close to or in retirement to protect them from unexpected drops in equities. Essentially, we are trading market risk for longevity risk early on by delaying the glide path roll-down process. In the opposite environment, we leave untouched the equity allocation of young participants, while decreasing it for other participants, with a maximum decrease reached close to or in retirement. Here, we are effectively trading longevity risk for market risk by accelerating the reduction in equity exposure. While both actions increase the glide path slope in non-neutral environments, the increase in sequence-of-returns risk is more than offset by the extra degree of market or longevity risk protection gained. The environment definition, a key component of our solution, centers around intermediate-term prospects of stocks versus bonds, as well as the level of market volatility. It results in a slow-changing signal, capable of catching long-term asset behavior while keeping turnover low. Furthermore, because such behavior is meanreverting, our signal forces a contrarian conduct on participants, which counters their momentum tendencies of buying high and selling low. Finally, to ascertain retirement success, we introduce our dashboard of metrics that should capture various risks faced by a participant through his or her lifetime in the product. We show that our solution improves on virtually all these measures. Ultimately, we believe our approach represents a distinct improvement over competing, more tactical, active glide path management mechanisms. The nature of our market assessment, calibration and frequency of positioning, and asymmetrical adjustment process are designed to seek smoother and more successful payoff streams for the greatest number of participants. 14

15 8 The distribution of wealth tends to follow a log-normal format, assuming portfolio returns are log-normally distributed. For log-normal distributions, mean-adjusted risk measures such as the coefficient of variation (CV) are more stationary than the traditional standard deviation. Our measure is essentially Sqrt(log(1+CV 2 )). Appendix DEFINITION OF VARIOUS METRICS The metrics we analyze for wealth at retirement (W 65 ) see Figure 9 for an illustration: Mean = average wealth, showing how much we can expect to have on average when we retire X Percentile Wealth = percentile 1-X of the distribution, showing what is the lowest outcome if we are to be among the best X% scenarios for wealth. We show values for X set to 99th, 95th, 90th, and 50th percentiles. Dispersion = mean-adjusted risk. We choose a more complex measure of dispersion than the typically used standard deviation. If portfolio average returns increase, without changing any portfolio risk parameters, both the mean and the standard deviation of W 65 increase. The standard deviation increase is misleading because it comes from something unequivocally positive, i.e., higher average returns. Hence, we choose a measure of dispersion that filters that effect out by essentially looking at the risk of W 65 expressed in logarithmic terms. 8 Dispersion 2 = ln( 1+ Var[W 65] ) ( = ln E[W2 65] E 2 [W 65 ] E 2 [W 65 ] Sharpe Ratio of Wealth = risk-adjusted return on W 65 ; it has the same concept as the Sharpe Ratio. We take the total increase in wealth from portfolio returns and set it against the risk taken as reflected by the variation of wealth at retirement. Following the rationale for Dispersion measure, we use logarithmic terms for wealth for this measure as well. SRW = ( ln ( E[W 65 ] t=26 65 Contribs t ) ) 2) -.5 * Dispersion / Dispersion For wealth distribution at various ages in retirement we compute a subset of the metrics above, namely the Percentile Wealth ones. Furthermore, we add Probability of Success measure, which is the percentile for which wealth is zero, thus showing the probability of having a positive wealth at that age. We deem a negative wealth as a retirement failure. ASSUMPTIONS To compute these metrics at retirement, we require a set of initial assumptions: Contributions: $1/year, inflation-adjusted, over 40 years. The exact value does not matter for our results as long as we keep them constant in real terms. Further, we tested our dynamic risk management effects with contributions growing over time, and the effects stayed qualitatively the same. Distribution of asset returns: American Century Investments assumptions, in inflation-adjusted terms. These values are in line with long-term market assumptions circulated by various providers, e.g., stocks have a 3.75%/year higher return than bonds, with a volatility of 15%, and a correlation between them close to zero. Returns are assumed to be distributed jointly log-normal. We tested non-normal distributions with both skewness and kurtosis, and the overall results about dynamic risk management remained similar. Allocations, or the glide path: the anchor glide path of our dynamic risk management solution. 15