Frequently Asked Questions
Why do MEPSIM's simulations begin in 2015?
MEPSIM uses 5500 data as the starting point for its simulations. Each year, every multiemployer plan is required to submit its 5500 data to the Department of Labor. But plans may wait up to 7 months after the end of a plan year to file their data. In addition, the data is not immediately available to the public because the Department of Labor must first process and compile it into a database. As a consequence, a complete dataset for all plans for year "X" is usually not available until year "X+2". Therefore, MEPSIM's simulations start two years in the past.

When computing the present value of assistance, does MEPSIM discount back to 2015?
Although 2015 data is used as the starting point for MEPSIM's calculations, MEPSIM computes the present value of assistance payments by discounting to July 1, 2017. Had we discounted back to 2015, the resulting present values would have understated the PBGC's obligations.

Why was 6% selected as MEPSIM's default rate-of-return on plan assets?
MEPSIM allows users to set the rate-of-return (ROA) at any level they wish. But when the model first loads into a browser, the ROA is initially set at 6%. We chose this level because, in our view, it represents a realistic long-run expectation for a 60/40 stock/bond asset portfolio. However, according to 5500 data, the median multiemployer discount rate is about 7.5%. This is high relative to our expectations.

While developing our ROA assumptions, we found the following capital market forecasts and analyses to be useful:

J.P. Morgan's "2017 Long-Term Capital Market Assumption Executive Summary"

Vanguard's "2017 Economic and Market Outlook: Stabilization, not Stagnation"

McKinsey Global Institute's "Diminishing Returns: Why Investors May Need to Lower Their Expectations"

Why was negative 1.5% selected as MEPSIM's default rate-of-increase of active participants?
MEPSIM allows users to set any assumption at any level they wish. But when the model first loads into a browser, the rate-of-increase of the number of actives is initially set to negative 1.5% for the next 10 years, and 0% thereafter.

Historical data reveals that, since the year 2000, the total number of active multiemployer participants has declined at an average rate of 1.5% per year. We used this value as our default short-range assumption. We were reluctant, however, to extend this assumption beyond the 10th simulation year, because it seems possible that a point-of-stability will eventually be reached. For the long-run, therefore, we have assumed a stable number of workers.

Why was positive 1.5% selected as MEPSIM's default rate-of-increase of contribution-per-worker?
MEPSIM allows users to set any assumption at any level they wish. But when the model first loads into a browser, the rate-of-increase of the contribution-per-active-worker is initially set to positive 1.5% for the next 10 years, and 0% thereafter.

Historical data reveals that, since the year 2000, contributions-per-worker have, in general, been increasing. At the same time, the rate of new benefit accruals per worker has remained roughly flat. As a result, workers and their employers pay more and more for the same benefit package. For some plans, this trend has led to a situation in workers could actually achieve a better benefit package if they simply invested their contributions in risk-free treasuries.

Therefore, we have assumed that gap between contribution increases and benefit increases vanishes after year 10 of the simulation.

Why was 2.55% selected as MEPSIM's baseline rate for discounting PBGC assistance payments?
In our opinion, the Treasury yield curve should be used to discount PBGC assistance payments. Using the curve from July 3, 2017, we computed the present value of projected assistance and, using this present value as a target, we determined that the level-equivalent discount rate is 2.55%. In other words, discounting with a flat rate of 2.55% produces the same present value as discounting using the entire yield curve. Therefore, as baseline, we set the rate for discounting projected assistance payments at 2.55%.

Why are MEPSIM's projected assistance payments in 2015 lower than actual payment levels?
MEPSIM projects plan insolvencies that are expected to occur in the future. MEPSIM does not contain any data for insolvencies that have already occurred -- that is, for the plans that are already receiving assistance from the PBGC. In FY2015, the PBGC paid-out about $100 million in assistance. MEPSIM's output table showing the projected stream of assistance excludes payments to plans that are already insolvent. However, for the purpose of determining the PBGC's date of insolvency, and for determining the required premium to prevent insolvency, we assume that the plans that are already insolvent require assistance payments totaling about $2 billion over the next 20 years.

Why isn't my pension plan in MEPSIM's multiemployer database?
There are approximately 1300 multiemployer pension plans, but only about 1200 of them are in our database. We excluded about 100 small plans from our database because they lacked sufficiently complete 5500 data to use in MEPSIM. The excluded plans represent about 4% of total liability computed across the multiemployer plan universe.

For the Central States pension plan, why does the projected contribution drop by $600 million between 2015 and 2016?
In general, MEPSIM uses a plan's reported total contribution in 2015 as the basis for projecting contributions forward in time. However, in the case of Central States, documentation contained in the plan's 5500 filing for 2015 indicates that the 2015 contribution was an upward spike that will not persist in the future. The spike was due, in part, to a significant withdrawal liability payment that took place in 2015. Therefore, we adjusted the projected 2016 contribution to elimate the effects of the 2015 spike.

Are MEPSIM's results accurate?
Any model tasked with simulating the entire multiemployer system Ė which is comprised of more than 1300 pension plans -- must depend to a large extent on 5500 data, because it is not possible to gather detailed data for each individual pension plan. 5500 data provides information about the aggregate state of each pension plan but does not provide granular information such as the distribution of plan participants and their accrued benefits by age and sex. Therefore, a model of the multiemployer system must squeeze the greatest possible predictive value out of the 5500 data.

To maximize the value of the 5500 data, MEPSIM uses the data to simulate the entire plan universe as opposed to merely a small subset of plans. Any subset of plans is unlikely to have the same distribution of demographic and financial characteristics as the universe as a whole. This bias could distort a modelís results. We avoid this problem entirely by simulating all plans rather than merely a subset of plans.

While MEPSIM circumvents the problem of sample bias, other challenges cannot be sidestepped. The two principle challenges which arise due to the lack of granular data are as follows: (1) at the outset of a plan simulation, how should the model create a stream of accrued benefits consistent with the plan's reported liability?; and (2) how can we estimate the fraction of a planís accrued benefits that will be covered by the PBGCís benefit guarantee?

In regard to issue (1), MEPSIM uses an algorithm to impute an initial stream of accrued benefits with a present value equal to the plan's reported liability. Roughly speaking, the greater the number of retirees relative to active workers, the lower will be the duration of the benefit stream created by the algorithm. This begs the following question: are MEPSIMís simulation results sensitive to adjustments of the algorithm? Our tests indicate that algorithmic adjustments Ė such as an increase or decrease of liability duration Ė have little impact on MEPSIMís aggregate results.

With respect to issue (2), the PBGCís benefit guarantee is expressed on a per-year-of-service basis. Therefore, to precisely calculate a participantís guaranteed benefit, one needs to know not only their monthly benefit but also their years of service. This information isnít available in the 5500 data. In fact, even the plans themselves might lack such data in electronic form for all participants. We dealt with this problem by using a planís normal cost per worker Ė which, by definition, captures the benefit accruals associated with one year of service -- as a means to estimate the covered-benefit-percentage applicable to the planís entire liability. By ďcovered-benefit-percentageĒ, we mean the ratio of the benefit that the PBGC will pay to the individualís total accrued benefit. Our methodology assumes that the covered-benefit-percentage of retirees is identical to that of todayís workers. For some plans, this might lead to an over-estimate of the covered-benefit-percentage, while for other plans it might lead to an under-estimate. The covered-benefit percentages generated by MEPSIM appear reasonable, but we do not yet have a good means to assess the accuracy of our approach.

While tweaks to the model's methology can shift its simulation results a bit, these shifts are small compared to the model's sensitivity to changes of user-specified assumptions. For example, decreasing plans' expected return-on-assets by 1% has roughly a 50% upward impact on the estimated total present value of PBGC assistance. Thus, it is the economic assumptions that are of central importance as opposed to the model itself. Any refinements to MEPSIM will have a small impact compared to the impact of assumption changes.

What is "normal cost"?
Normal cost is equal to the actuarial present value of benefits accrued over the course of a year. For example, a plan might have a liability of $100 million and a normal cost of $5 million. $100 million is the present value of all outstanding benefits, while $5 million is the annual rate at which additional benefits are accrued by todayís active workers.

What is "duration"?
Duration measures the length of time, in years, across which a set of future cash flows stretches. More precisely, it is the weighted average of the time-until-payment (in years) of the cash flows. For example, a liability (or asset) consisting of a single payment to be paid immediately would have a duration of zero. A liability consisting of two equal payments to be made one year in the future and three years in the future would have a duration of two years. The duration of pension liabilities typically ranges from 10 to 20 years. In general, the greater the ratio of pensioners to workers, the lower will be the duration of pension liabilities.