Home > Stochastic Forecasts

Stochastic Forecasts

See also Stochastic Forecast Output.

Stochastic forecasts show the financial status of the pension plan over a specified future period based on a stochastic economic environment. In effect, a stochastic forecast consists of many separate deterministic forecasts, where each deterministic forecast is selected on a quasi-random basis.

Stochastic forecasts run very quickly because each forecast is based on one or more previously selected Core Projections, for which a large number of actuarial factors used in the forecast process have been stored.

Stochastic forecasts interpolate baseline and sensitivity Core Projection Output values using logarithmic interpolation (which is especially beneficial for computing accurate extrapolated values “in the tails” of distributions of forecast results). For details about the interpolation method, see the Technical Reference article entitled Interpolation of a Core Projection's results.

If you include more than one Core Projection in a Stochastic Forecast, please be sure that the parameter settings are consistent among the Core Projections with respect to (in particular): Valuation Date, liability methods (including timing of employee contributions), funding valuation interest rates, accounting valuation interest rates and, in the U.S. qualified pension mode, applicable law. Thus, for example, if you include (pension mode) accounting Cores with different ABO interest rates or (U.S.) PPA funding Cores with different valuation interest rates (such as spot rates for one Core Projection and segment rates for another one), the forecast run will abort.

The items requested in the Stochastic Forecast dialog box are as follows:

Name is a descriptor for the forecast currently being defined. You are encouraged to provide as complete a description as possible so that you can easily keep track of the various forecasts that are run.

Core Projections lists the (baseline) Core Projections included in the forecast. Additional Core Projections can be added to this list by clicking the Add / Omit button and selecting from the available (and not previously referenced) Core Projection library entries. A Core Projection can be deleted from the list by clicking its name and deselecting it under the Select items to include parameter. (The List parameter allows you to display “All” Cores unhidden in the current Project or just the “Completed Cores” (“Completed Cores (Ongoing Liability)” in U.K. mode), that is, executed and saved Cores.

Note that if more than one Core Projection is included for the same database records, it is important that future experience (with respect to assumed actual decrements, salary increases, increase rates, benefit payments and payment forms) be the same. Otherwise, it is likely that an “inconsistent demographics” message will appear at the start of the forecast run and, in order to be sure the results are as intended, the underlying cause(s) should be addressed before proceeding. In particular, if benefit payment amounts involve lump sum factor values that differ between (separate) funding and accounting Cores, then future experience may be inconsistent (between funding and accounting) and a warning will be generated.

If you include more than one Core Projection in a Stochastic Forecast, please be sure that the parameter settings are consistent among the Core Projections with respect to (in particular): Valuation Date, liability methods (including timing of employee contributions), funding valuation interest rates, accounting valuation interest rates and, in the U.S. qualified pension mode, applicable law. Thus, for example, if you include (pension mode) accounting Cores with different ABO interest rates or (U.S.) PPA funding Cores with different valuation interest rates (such as spot rates for one Core Projection and segment rates for another one), the forecast run will abort.

In all modes except OPEB and U.S. public pension, the Overrides button provides a way to override, or replace, certain liabilities and costs calculated in the baseline Core Projection(s) with liabilities and costs as calculated by other Cores. The variable output values that may be overridden are:

For example, to override all PPA Target Liabilities (U.S. qualified mode) generated from the Core Projection(s) specified in the Included Core Projections, click the Overrides button, choose PPA Target Liabilities and select the Core Projections(s) that contain the appropriate values for PPA Target Liabilities. The total PPA Target Liabilities will be overridden for all forecast years and replaced with the aggregate PPA Target Liabilities produced by the alternate Core Projections(s) selected under the Overrides button. Although the total number of alternate Core Projections required to override the specified liability may vary from the total number of Core Projection(s) specified as Included Core Projections, the total population represented should remain constant; otherwise, ProVal will generate a warning that the demographics are not consistent among the Core Projections.  In the special case that there are no records mapped to the ProVal active status, or no records mapped to a ProVal inactive status, in the baseline Core Projection(s) but there are such records in the override Core Projection(s), ProVal will issue a warning and will not process active liability overrides, or inactive liability overrides, respectively (and thus there will be no active, or inactive, records, respectively, reflected in the Stochastic Forecast results).  

A check mark on the Overrides button indicates that at least one Core Projection has been selected for at least one type of liability. Note that, for the selected liability/liabilities, ProVal will override the results in total (with the exception of the special case already mentioned, of no active participants, or no inactive participants, in the baseline runs). Thus if a participant record is present in the database(s) used for the baseline Core Projection(s) but not present in the database(s) used for the override Core Projection(s), ProVal will not use the baseline results for that record in the final results (after the override is applied). Thus, to be sure use of override Cores produces the desired liability values, you must maintain a one-to-one correspondence between a record in the baseline run and another record in the override run.

Note that, with respect to assumed actual decrements, salary increases, increase rates, benefit payments and payment forms, it is important for future experience to be the same in the override Core as in the baseline Core to which it corresponds. This also applies if more than one override Core Projection is included for the same database records – as, for example, if separate override Cores are specified for different liabilities. If ProVal detects that future experience is not the same among the baseline and override Cores, an “inconsistent demographics” message will appear at the start of the forecast run and, in order to be sure the results are as intended, the underlying cause(s) should be addressed before proceeding.

In the Canadian registered mode, a particular instance of inconsistent demographics can occur if Core Projections are used to override solvency liability: It is recommended that you use the same Plan Definition for an override Core Projection as used for the corresponding baseline Core Projection, although this means that death and disability Benefit Definitions are defined for the solvency (override) Core Projection(s). Then use post-decrement probabilities (of 1’s and 0’s) to turn on and off the benefits applicable to the baseline and solvency (override) runs, respectively. This ensures that the solvency (override) valuations at future forecast valuation dates (after the baseline valuation date) include new inactive participants who emerged the preceding year from active participants decrementing because of death or disability. That is, if a participant dies between the baseline valuation date and a future valuation date and leaves a beneficiary, or if the participant becomes disabled during that period, ProVal needs Benefit Definitions initiated by death and/or disability in order to value the beneficiary or disabled participant as a (new) inactive member.

Future Events lists any assumption changes (e.g., a change in the valuation mortality rates) which occur at a future valuation date. An event consists of one or more core projections, to be used to replace the results of the baseline core projections, along with the date the replacement should occur.  Note that valuation interest rate changes are automatically captured as an assumption change. Therefore, do not set up an event to capture changes in the valuation interest rate.

Use the Add Event button to enter the parameters for your event.  Use the split button arrow to select either a Blank Event or Same as Baseline. Selecting Same as Baseline will initialize the event with all the Core Projections (including Overrides) specified in the Baseline event.  This is useful for complex setups with many overrides where only a single Core Projection needs to be changed from the baseline. Once in the event dialog box, you will be required to enter the year in which the event occurs and to select at least one Core Projection. Select the Type as Gain or Loss or Assumption change to determine how the event should be treated for purposes of determining contributions and expense. (In pension modes, plan changes should be handled through the Plan Amendments feature of Projection Assumptions). Events will be effective on the valuation date containing the event year (i.e., events only happen at beginning of year) and the event will continue for the duration of the forecast (unless there is a subsequent event). The same liability override options available in the baseline are available for events.  You may have multiple events in your forecast, but only one event may be coded for a particular year. When an event is coded, all available output (demographic, liabilities, benefit payments, etc.) will reflect the results from the Core Projection(s) specified as part of the event.  To edit an event, you may either double click directly on the event or click the Edit button. Use the Erase button to remove an event from the list.

The Apply Scaling factors box becomes accessible only if scaling factors were specified for one or more of the referenced Core Projections. If the Apply Scaling Factors checkbox is selected, the scaling factors specified in the Core Projections will be applied to the results before calculating plan contributions and expense.

Select the Stochastic Assumptions from the library of Stochastic Assumptions unhidden in the current Project. Select the appropriate assumptions for the forecast to be performed, or click the button to create a new library entry.

Select the Asset & Funding Policy from the library of Asset & Funding Policies unhidden in the current Project. Select the appropriate Asset & Funding Policy for the Deterministic Forecast to be performed, or click the button to create a new library entry.

Enter the number of Forecast years (from 1 to 100, inclusive) you wish to run. You can run a forecast for a period up to, but not in excess of, the shortest projection period of any Core Projection referenced in this forecast library entry.

Enter the Number of Trials (stochastic trials) you wish to run for each year in the forecast period. You can select any number between 10 and 25,000. Generally 1,000 or more trials are run.

Click the Trial Detail button to capture and store variables on a trial by year basis. This is very helpful in analyzing and diagnosing the results obtained, to see that, overall, results are reasonable and parameters are set correctly. In the Trial dialog box, check the Write stochastic output by trial to a file box to save trial detail to an external csv file or an Excel spreadsheet.  If an Excel spreadsheet is chosen, then the results for each asset mix will be saved to a separate tab. The remaining parameters, including lists of standard output variables and custom output variables selected behind the Custom Output button, then become accessible:

In the Save as field, enter the name (and, optionally, the path) for the file to be created to contain the trial detail results. By default, the file will be saved in the client directory. The Browse button can be used to navigate to a particular folder or existing file. Note that if you reference an existing file, you will be asked (during execution) if you would like to Replace the existing file or whether you would like to Append the new results to the existing file.

Select one or more Variables from the list of available output variables.  Note that a variable labeled "inactives" refers to both initial and emerging inactive members.  

If you have created and selected any Custom Variables, you may select one or more from the list of custom variables unhidden in the current Project.

Select one or more Asset Mixes from the list of Asset Mixes.

Select one or more forecast Years.

The dialog box will show, to the right of the Variables parameter, a summary of the file contents to be created, based on your selections for output variables, asset mixes, years and trials. Be careful to limit the data to what is actually needed, because the resulting file can become very large and unwieldy if too many results are requested.

A check mark on the Trial detail button indicates that stochastic output by trial will be written, for at least one forecast year, to an external file.

The Custom Output button allows you to select one or more previously-defined Custom Variables to include in the Stochastic Forecast. Including them during execution allows you to review (under the Output menu and behind the View button of this forecast) the output associated with each Custom Variable. Note that you cannot select a custom variable for trial detail until you have selected it behind the Custom Output button. A check mark on the Custom Output button indicates that at least one Custom Variable has been selected. Also set the ETL Confidence Level percentile. This defines the worst-case tail boundary point used in the ETL calculation. For a given variable, the ETL is the average value in the worst-case tail of the distribution. ETL can be selected as an output percentile.

When you have completed all the inputs to the forecast, you should click the Replace or Save As New button to save the parameter settings. Then you can reenter this dialog box and run the forecast by clicking the Run button.

The Run button checks to see that all parameters have been set and then, if no errors are found, executes the forecast. An indication of what steps are being performed, as well as percentage complete and time elapsed, will show on the screen during execution.

When the forecast is complete, a message will appear describing how long it took to run, as well as warning messages and other processing messages. These messages can be printed, or viewed at any later time by means of the View button.

 

Viewing Output

The View button displays summary output, along with inputs and processing messages. Click the Options button to:

Alternatively, you can create customized output and view stochastic trial exhibits by using the Output pane (on the right side of ProVal’s main window).