Nobel laureate Bill Sharpe has focused the most recent phase of his career on retirement income analysis. A big part of that work has been the creation of a 21-chapter online financial economics textbook covering the various subject areas related to retirement planning. What I found most interesting in studying the text were Sharpe’s observations about current retirement planning practices – especially his skepticism about products and approaches used by advisors. I’ll provide a review of the text, plus additional comments on these views.
The textbook is titled RISMAT, which stands for Retirement Income Analysis (with scenario matrices). It covers subjects within the scope of financial economics such as utility theory and the valuation of assets, and others from outside such as demographics and building mortality tables. Incorporated in the text is a comprehensive system for retirement planning developed by Sharpe using Matlab software. I was surprised to discover that Sharpe is not a retired professor who dabbles in computer programming, but a skilled and experienced programmer capable of building large, integrated and full-featured systems.
For those familiar with Matlab, Sharpe’s textbook chapters show the development of the Matlab code for the various modules in the system. To run the system, one has to have access to Matlab software (requiring a pricey subscription) and experience working with Matlab. However, for non-Matlab users, the text provides an easy-to-understand descriptions of the program structure.
Sharpe’s general approach to retirement planning analysis reflects his extensive knowledge of financial economics and provides a quite different perspective than the research one encounters in publications that planners and advisors will be most familiar with, such as the Journal of Financial Planning. The good news is that Sharpe offers a refreshing new way of looking at retirement planning analysis; the bad news is that reading his textbook can be challenging, particularly for those like me without a strong background in financial economics. But, based on my own experience studying the text, it’s well worth the effort – even if a few of the chapters turn out to be difficult. I found the text to be a pleasure to read; Sharpe is articulate and precise in his language, but also displays a wry sense of humor.
To learn more about Sharpe’s research and ideas, I also recommend Michael Finke’s February 2019 article in Investment Advisor. An older piece I recommend is the 2014 Advisor Perspectives interview Sharpe did with Bob Huebscher.
Software structure
The “Mat” in Matlab refers to matrix, and RISMAT is built on a matrix structure. One can think in terms of a matrix with 100,000 rows, each representing a separate retirement scenario, and 60 or so columns representing the year-by-year progression of retirement to the maximum possible length. Typically each of the rows would be generated by some type of Monte Carlo process.
Throughout the text, Sharpe uses an example of a retired couple, Bob 67 and Sue 65. The first matrix he builds is a personal states matrix for the couple. The personal states are: both alive, Sue only alive, Bob only alive, and both deceased. These personal states are generated year-by-year for each row of the matrix by applying random number generation to a mortality table to appropriately recognize the random nature of longevity. We end up with 100,000 rows (scenarios) with different longevity characteristics in terms of the length of retirement with both alive, followed by differences in who survives and for how long.
The retirement modeling then operates by combining matrices. For example, let’s say Bob and Sue begin retirement with $1 million, invest all their savings in a 60/40 stock/bond portfolio, and take annual withdrawals based on required minimum distributions (RMDs). RISMAT will generate 100,000-row matrices of year-by-year investment returns using Monte Carlo techniques and the RMD withdrawal percentages will be applied to year-by year savings balances. The personal states and investment matrices will be combined to produce a matrix that shows, for each of the 100,000 scenarios, year-by-year withdrawals and bequests after the last member of the couple dies. If there are investment or advisor fees, these will also be built into a matrix so that the scenario matrices can be generated both before and after fees.
Lockboxes
Besides the matrix approach, another feature employed in RISMAT is the use of lockboxes. For example, it would be feasible at the start of retirement to build a product that sets up a lockbox for each future year’s retirement income. A “safe” lockbox for year 20 of retirement might contain $10,000 invested zero-coupon Treasury inflation-protected securities (TIPS strips) maturing at year 20. If the yield is 1%, the lockbox would provide $12,202 in real dollars at year 20. A riskier approach could be taken by placing a market portfolio of stocks and bonds in the lockbox, so that the result after 20 years would be variable. However, lockbox amounts depend only on cumulative returns – the sequence does not make a difference, avoiding what Sharpe refers to as path dependency. In the various scenarios, the balances in unopened lockboxes after the death of the last member of the couple become the bequest or the estate.
Sharpe also discusses lockbox annuities where an insurer funds the lockboxes based on assumed mortality and takes the risk that mortality turns out better or worse than what was assumed. These annuities could provide either fixed payments (real or nominal) or variable payments depending on the types of investments placed in the lockboxes.
The combination of matrices and lockboxes contributes to an overall structure that is easy to grasp at a conceptual level.
Analysis
This general setup theoretically allows individuals to compare outcomes from different strategies. However, it would obviously be impossible for any human to compare two matrices with 100,000 rows and decide which they like best. There is a need for summary information such as graphs showing average and perhaps 10th percentile yearly income for the two approaches. Sharpe recognizes that this could get into the realm of coming up with the best data visualization approaches for comparing alternatives. Another way of comparing matrices of outcomes could be to apply utility measures like certainly equivalents. However, Sharpe indicates that such an approach, while useful for research, might be too “black box” for most consumers, and some form of data visualization could be more appropriate.
RISMAT does produce charts that show the present values of how initial savings on average get split between providing for a couple when both are alive, providing for either when widowed, providing funds for the estate, and amounts spent on fees. Sharpe feels it is useful to show actual amounts going to fees of all sorts rather than just showing outcomes on an after-fee basis.
Investments, products and strategies
In modeling investments, Sharpe uses an approach based on the capital asset pricing model (CAPM), which posits that the optimum relationship between return and volatility can be achieved with portfolios that combine a risk-free asset and a highly diversified market portfolio that includes proportionate shares of all publicly-traded investible assets. Sharpe develops a proxy for the market portfolio based on a mix of four Vanguard funds:
- Total stock market index
- Total bond market index
- Total international stock market index
- Total international bond market index
He uses the Vanguard inflation-protected securities fund to represent the risk-free asset.
His approach assumes, “that markets are sufficiently efficient that it is difficult to identify securities or classes of securities that are habitually underpriced or overpriced.”
Views on advisor practices
Sharpe raises questions about approaches that he sees advisors utilizing in making retirement planning recommendations. He takes a macro or complete-markets view of the investment world, which could be crudely summarized as follows: If someone is outsmarting the market, someone else must be “out-dumbing” it, and why would they keep doing that? Quoting Sharpe: “It is important to emphasize that if one investor chooses increased exposure to a factor relative to the market, some other investor must accept decreased exposure relative to the market. And they cannot both beat the market. Over any given period, some tilt strategies will do better than the market, while others will do worse than the market.”
He focuses on efficiency in building portfolios of investments and financial products. Portfolios would be considered inefficient to the extent that a portfolio could be built at a cheaper cost that would generate the same probability distribution of future outcomes. Sharpe applies valuation measures to calculate the amount of inefficiency in various types of portfolios. A common source of inefficiency is deviating from a market portfolio.
Here are some examples where he raises questions about advisor practices:
Rebalancing
Many advisors recommend periodic rebalancing between stocks, bonds, and perhaps alternative assets. Sharpe views such rebalancing as making an active bet against the market portfolio. He also takes issue with glide-path strategies, because they rely on rebalancing to a pre-determined asset allocation.
Active bets
Many advisors recommend placing higher-than-market weightings on certain asset classes that they view as undervalued. Sharpe, again taking a macro view, argues that this necessarily requires that others need to be under-weighting those undervalued securities. Sharpe is clearly skeptical about any strategies that attempt to outguess the overall market – again reflecting his macro perspective. Whatever someone sells, somebody else has to buy.
Fixed withdrawals
Sharpe is not a fan of fixed withdrawals (e.g., the 4% rule) where the year-by-year withdrawals are independent of what’s happening to the underlying investment portfolio. Quoting Sharpe, “Speaking for myself, and putting the subject in non-academic terms: it just seems silly to finance a constant, non-volatile spending plan using a risky, volatile investment strategy.”
Annuities
Sharpe recognizes the unique benefit that can be provided via mortality pooling, but isn’t passionate for or against annuities. But he notes that, “without annuities, one has to accept a tradeoff. The smaller the chance of running out of money, the greater the likelihood that a substantial amount of money will be left unspent, then provided to an estate.” He does, however, raise questions about annuities with guaranteed lifetime withdrawal benefits because of the variability in real income generated.
Further comment
Developing an appreciation for Sharpe’s macro focus and CAPM-based investment philosophy raises questions about many of the retirement planning strategies that advisors employ. A random sampling of such approaches from my recent reading includes:
- Alternative asset classes instead of stocks
- Bucket strategies – near-term, medium-term, and long-term buckets
- Bond tents – higher bond allocations around the time of retirement
- Withdrawals from bonds after years when the stock market performs poorly
- Guaranteed lifetime withdrawal features of fixed-index annuities and variable annuities
- Stocks that pay high dividends as an income source
- Higher risk bond portfolios to generate more cash flow
Some of these strategy recommendations are based on nothing more than their sounding good, while others may be based on research that doesn’t fully appreciate the constraints imposed by a macro view of markets. Perhaps there are a few strategies that, indeed, take advantage of persistent market anomalies, but these represent a small minority. Sharpe’s macro view certainly provides a basis for asking a lot of questions.
Advice
The concluding Chapter 21 of the text on advice is worth a careful reading. It succinctly summarizes many of the key ideas presented in the preceding chapters. I’ll conclude by quoting two passages from from this chapter:
“If there is any conclusion to be reached after reading the prior twenty chapters it is this: comprehending the range of possible future scenarios from any retirement income strategy is very difficult indeed, and choosing one or more such strategies, along with the associated inputs, seems an almost impossible task. At the very least, retirees will need some help. Enter the Financial Advisor. Ideally, he or she will have a deep background in the economics of investment and spending approaches, sufficient analytic tools to determine the ranges of likely outcomes from different strategies, and an ability to work with clients to find approaches that are suitable, given their situation and preferences. Moreover, the amount charged for providing such advice should be well below its added value. Tall orders indeed.”
“It is the author's belief that an understanding of the issues covered in this book, an ability to make useful projections of the ranges of possible future outcomes and their properties, plus the ability to effectively communicate such projections to a client could significantly enhance the ability of a Financial Advisor to help clients make appropriate retirement income plans.”
Joe Tomlinson is an actuary and financial planner, and his work mostly focuses on research related to retirement planning. He previously ran Tomlinson Financial Planning, LLC in Greenville, Maine, but now resides in West Yorkshire, England.
More Factor-Based Investing Topics >
