RiXtrema Quantitative Research. Retirement Plans Are Wasting $12.32B in Fees: How Are Your Clients & Prospects Doing?

Transcription

1 RiXtrema Quantitative Research Retirement Plans Are Wasting $12.32B in Fees: How Are Your Clients & Prospects Doing? 2016

2 Significant Fee Waste in Retirement Plans New Study Using Quantitative Methods in 401kFiduciaryOptimizer Summary: It has been known for some time that fee inefficiency is pervasive in 401(k) plans. Pressure has been building up on retirement plans to address this problem and with the new DOL Fiduciary Rule this pressure is now applied to the whole industry of financial advice. However, there has been surprisingly little quantitative evidence to show how much retirees are overpaying in 401(k) plans. We analyzed 7472 retirement plans from the DOL EFAS database. Using quantitative methods we estimate that participants could save on average.44% a year by switching into lower cost investments that are quantitatively very similar to those they already hold. Similarity is defined as a combination of category filters, together with historical and forward looking predicted correlation based on a multi-factor model. otal 401(k) plan assets held in funds stood at $2.8 trillion at the end of his means that savings of approximately $12.32 billion annually are being wasted in the retirement industry. Assuming 6% average return and net inflow equal to 10% of plan assets, this will compound to almost $90 billion wasted over 5 years! 401(k) plans in the US are tremendously inefficient; however, the awareness of this problem is growing. Only those advisors who are able to respond to this challenge will end up surviving the upheaval that will result from these trends. Introduction It has been widely argued that 401(k) and other retirement plan participants are poorly served by plan menu choices filled with expensive mutual funds. Many (but not all) of those expensive funds tend to underperform over time due to higher fees and lack of consistent alpha. 401(k) plans held $2.8 trillion of fund assets at the end of 2014 and a great deal is at stake for participants in this debate. However, there is surprisingly little quantitative evidence regarding fee inefficiency in retirement plans. Estimates of waste in plans vary; for instance the DOL used an average figure of 11.3 basis points as an estimate of waste in 401(k) plans. It is not clear where that figure came from and it has been widely criticized as a result. 1 Our ultimate goal is to find a low fee replacement for high fee funds, but only where we can prove that the replacement is not materially changing the risk/return profile offered to participants in the current menu 2. Our goal is to identify those high-fee funds that exhibit a great degree of similarity to low cost alternatives, measure that similarity and offer evidence-based low fee replacements. his will give us 1 Assessing the Department of Labor s Assumption that 401(k) Plan Participants Pay Fees that are Higher than Necessary, Investment Company Institute the question that is not solved by this study.

3 the ability to estimate potential savings to retirement plan participants from this quantitative rebalancing. We will define similarity more precisely later in the text, but generally we define similarity as a correlation between two funds subject to condition that funds are in the same category. Materials & Methods: Plan Data in 401kFiduciaryOptimizer Plan Screener We have studied 7472 retirement plans from the DOL EFAS database. Certain 401(k) and 403(b) plans have to submit a long form 5500 which includes a schedule of assets, also called Schedule H. Only plans for which more than 80% of the investments reported in Schedule H could be identified are included in 401kFiduciaryOptimizer Plan Screener for advisors who want to present this information as part of their proposal to the plan sponsor. Materials & Methods: Fund Data in 401kFiduciaryOptimizer For calculation of Similarity we have the following building blocks. - History of returns for 31,589 funds used for calculation of historical correlations (source is Bloomberg, Yahoo and Quandl API) - Fund objective (source is Bloomberg) - Position data for mutual funds and EFs (from SEC EDGAR and EF provider websites respectively) - Multi-factor multi-asset class factor model used to forecast correlation for those funds where holdings are available Materials & Methods: Similarity Calculation In order to calculate potential savings from switching to lower cost alternatives, we need to select an investable universe that plan menus can choose from. We assume Open Architecture platform with 31,589 funds. Our next goal is to calculate Similarity measure which we will use to select the most similar lower expense fund to replace a more expensive incumbent fund. Definition of Similarity: We define Similarity between any two funds as: H P Similarity 2 Where:

4 P - Projected correlation based on fund s holdings and use of multi-factor model (where holdings are available, for more detail see Appendix A for holdings data preparation and Appendix B for projected correlation calculation) H - Historical correlation based on past returns (for more detail, see Appendix C) Materials & Methods: Replacing Expensive Funds For every fund that is currently in the plan, we find all funds with the same Bloomberg Objective 3. We then find that satisfy the following criteria: -. 9 H - Similarity. 9 Some funds do not have holdings data, and only historical correlation is used for them. We sort the results of these filters in the ascending order by the Net Expense ratio to find the lowest cost similar funds. he logic behind these conditions is relatively straightforward. We are looking for funds that would be qualitatively (category) and quantitatively (past and projected correlation) similar to the fund in the plan. As will be shown, many incumbent high fee funds will have lower fee alternatives with similarity greater than.9 and will replicate the more expensive fund very closely. Materials & Methods: Calculating Participant Savings Any fund in the plan with at least one lower cost similar alternative can be replaced with that alternative without significantly changing risk/return profile of the plan menu. he difference in net expense after accounting for assumed revenue sharing represents savings for participants 4. hus, the plan savings from switching to low fee funds with high Similarity are as follows: Given: xh - expense ratio of the high expense fund minus the 12b-1 fee 3 Using Morningstar Category does not make any substantial difference. 4 here is one more important point that needs to be accounted for in order to calculate the savings correctly. Many plans, especially smaller ones, employ revenue sharing arrangements. hese arrangements, while increasingly frowned upon due to their opacity, are still legal. However, if we are switching to lower expense fund, likely an index fund, there will be no revenue sharing. We have to account for that by approximating revenue sharing, since it is not reported clearly on form We assumed that revenue sharing is equal to 12b-1 fee of the fund. his is a reasonable assumption, though in many cases plans or third-party administrators will negotiate special revenue arrangements that could be better than the 12b-1 fee.

5 xl - expense ratio of the low expense fund % P / A - is otal Plan Savings expressed in percent per annum % P / A Where: Z f 1 w f *( x H x ) L Z - total number of funds in the plan Exhibit 1 Plan Savings For a Sample $18.35M Retirement Plan From 401kFiduciaryOptimizer Exhibit 1 contains an example of a retirement plan savings calculator with $18.346M in assets with assumed 10 year horizon, 6% average return and $1,8M annual inflow. he green button shows the overall savings of $2,183,837 which is equivalent to 54 basis points per annum. he screenshot contains only one plan fund Invesco Real Estate, along with low fee alternatives which have Similarity of close to 1.he overall savings calculator in the upper right assumes that the plan switches from each incumbent fund to the lowest fee similar fund (in this case Schwab US REI EF), but only if such alternative exists based on the similarity. Users can choose another alternative, say DFA Real Estate Securities. Results Exhibit 2 below presents the results of our analysis for the overall universe as well as plans stratified by size into buckets.

6 Exhibit 2 % Per Annum Savings From Switching o Lower Expense Funds With High Similarity Overall >100M 50M to 100M 10M to 50M 1M to 10M <1M Count Average %P/A 0.44% 0.33% 0.35% 0.41% 0.46% 0.49% Standard Deviation %P/A 0.20% 0.16% 0.16% 0.19% 0.20% 0.22% Median %P/A 0.46% 0.36% 0.38% 0.44% 0.48% 0.51% 75th Percentile %P/A 0.56% 0.46% 0.47% 0.54% 0.57% 0.61% 25th Percentile %P/A 0.30% 0.19% 0.23% 0.27% 0.32% 0.34% Exhibit 3 Histogram of % Per Annum Savings For Universe of 7472 Plans

7 Exhibit 4 - Histogram of % Per Annum Savings For Plans >$100M Exhibit 5 - Histogram of % Per Annum Savings For Plans Between $10M and $50M

8 Discussion Overall results suggest that for an average plan the available savings from quantitative rebalancing are forty-four basis points per annum. his means that approximately $12.32B is wasted in retirement plans annually. Assuming 6% average return and net inflow equal to 10% of plan assets, this will compound to almost $90 billion wasted over 5 years! As expected, larger plans are somewhat less wasteful with a mean and median for plans with assets greater than $100 million being respectfully.33% and.36%, while the same numbers for plans is between $1-$10 million are.46% and.48%. But given their size and bargaining power, they could do much better in expense efficiency. It is worth noting the long right tail of the histogram in Exhibits 3 and 5. It appears that there are some plans in the smaller buckets that are egregiously inefficient with some exhibiting available savings well in excess of 1%. 401(k) plans in the US are tremendously inefficient; however, the awareness of this problem is growing. Only those advisors who are able to respond to this challenge will end up surviving the upheaval that will result from these trends. References: Kwak, James Improving Retirement Savings Options For Employees. University of Pennsylvannia Journal of Business Law, 15 (2013), pp

9 Ayres, Ian, Quinn,Curtis. Beyond Diversification: he Pervasive Problem of Excessive Fees and Dominated Funds in 401(k) Plans. he Yale Law Journal, Vol. 124, No. 5 (2015), pp Appendix A: Holdings Based Data For Funds Holdings data for mutual funds comes from forms N-Q and N-CSR (for EFs from the provider websites). hose forms are parsed from the SEC EDGAR website and Microsoft Full ext Search (SQL Server) is used to map security names to tickers for equity funds. Fixed income instruments are recognized by the issuer, coupon, maturity which are present in all EDGAR forms. Every instrument is then added to commercially available RiXtrema Multi-Asset Class Risk model. Equities are added by stepwise regression to calculate sensitivities to factors like equity beta, liquidity, size industry. Fixed income instruments are added via pricing mechanisms to calculate sensitivities to different points on the yield curve and sector/rating buckets for credit risk. Out of 31,589 funds in our universe, 21,288 funds have holdings information that was readable. Appendix B: Forward Looking Correlation Our goal is to calculate correlations between two funds to determine how similar they are, but do it on a position level. he reason we need to calculate this is that for some funds historical returns are not a good indicator of behavior, because they change their profile often. It is also important to calculate position level correlations for fund without long history. Let us define: W - weighted average exposure of the fund to a risk factor (for example, to equity beta, effective duration, liquidity etc.) 5 W w* H Where: w - vector of weights of each security in a fund H - matrix of exposures of all securities to each risk factor in a multi-factor model (number of securities by number of factors) 6 SD(A) - annualized forecasted standard deviation of the fund 5 For a full list of factors and risk model calculations see RiXtrema GML Whitepaper 6 Risk model is based on Arbitrage Pricing heory and advanced extension of Modern Portfolio heory, for more detail read here:

10 G - is the idiosyncratic relationship matrix for fund A vs. fund B. It has number of columns equal to number of positions in fund A and number of rows equal to number of positions in fund B. Each element of the matrix has either zero if a security listed in a given column is different from the one listed in a row or else it has the idiosyncratic standard deviation 7 of the security. For example, in row 4 and column 12 of the matrix we have the intersection of IBM and MSF. Since these are different stocks, the element in the matrix will be zero. Another element, say row 4 and column 8 has IBM for both. In that case the element of the matrix will be equal to idiosyncratic risk of the IBM. Holdings Based Correlation (Fund A vs. Fund B) = W A * C * WB wa * G * w SD( A) * SD( B) B Appendix C: Historical Correlation Historical correlations are calculated as a standard Pearson product-moment correlation coefficient. he formula for correlation between funds A & B is: A, B i1 i1 ( r ( r A, i A, i r ) * A A r ) 2 * i1 ( r i1 B, i ( r r ) B, i B r ) B 2 Where: r A, i - return of a fund A at time i ra - average return of fund A across time - number of periods in the sample 7 Idiosyncratic risk or idiosyncratic standard deviation is standard deviation of any asset (particularly a stock) that cannot be explained by systematic risk factors included in the model. his is a standard way to model investment risk.