Version 5.16: Performance Improvements and a Real-World Case Study

INTRODUCTION

In the most recent release (5.16) to the application, IMPLAN introduced new efficiencies in impact analysis processing, including Multi-Regional Input-Output (MRIO) projects. This article will discuss the new changes, run-time improvements, and features discussion with Dr. Patrick Scott of Louisiana Tech University about the projects he has been working on and his experience with the newest release.  

PERFORMANCE IMPROVEMENTS

In order to assess the improvement in project run-times, we ran a series of Projects through the new version of the application and compared them to those same Projects that were run through the previous version. This first comparison uses scenarios that include one or two Groups (Regions) with varying amounts of Events. These Projects were run without checking the MRIO checkbox.

As you can see, the new release provides run-time improvements over all of the scenarios, and the chart shows the increased performance improvements as projects get larger. The ‘New’ version averages about half the run-time of the ‘Previous’ version, and there are increasing returns to scale as projects increase in size. Obviously larger projects will have longer run-times as a general rule, but these statistics show measurable decreases in run-times. 

Now, let’s take a look at some MRIO projects! These projects all contain multiple Groups (Regions) and all of these Projects were run as MRIO Projects.

Again, the New version shows decreased project run-times in every scenario. As with the non-MRIO projects, we see an average run-time for the New version that is about half that of the Previous version, or roughly a 2X improvement for the projects which successfully completed in both versions. However, in the third and fifth scenarios the Previous version was not able to successfully run all of the events in the scenario on the first try. The New version ran these Projects with no issue. This shows that not only is the system processing existing projects in roughly half the time, it is simultaneously increasing the system’s maximum project size and the consistency with which projects are run successfully.

In the real-world example discussed below, Dr. Patrick Scott commented that “the efficiency gains from the update are quite impressive. I am curious to re-run some other projects to see how long they take now.”

PROCESSING CHANGES

Ok, now that we’ve covered some performance statistics, let’s talk about the actual changes and what’s different. Let’s start by doing a quick, high-level review of how a Project is processed. IMPLAN takes each Event that the user puts into a Group and creates a Final Demand (FD) vector that represents that Event. By multiplying the FD vector by a matrix of multipliers for the corresponding Region, IMPLAN arrives at estimates of economic impacts. Like so:

While the details can get a bit more complicated, the important part here is to understand that prior to this deployment release for each Event in each Group in a Project, IMPLAN needed to do the multiplication of the FD vector with the multipliers, which is a cumbersome calculation. So, if you had 2 Regions in your project and each Region had the same 2 events, then you would have needed to run 4 analyses (2 Regions X 2 events each = 4 analyses). That would look something like this:

In order to increase calculation efficiency, the new release places each of the Events in a Group as a column in a FD matrix. This means that multiple events can be done using a single multiplication step rather than each event needing its own, thereby reducing the number of total analyses needed for the project. So, the new version of the example above, with 2 Regions each, with Events 1 and 2, only requires 2 analyses, rather than 4. It looks like this:

This new method allows for many events to be processed inside of each individual analysis, making the system more efficient with each analysis and requiring fewer analyses per Project on average. This will save you time in running larger numbers of events and Regions, especially within your MRIO Projects.

CASE STUDY EXAMPLE

To find out more about how these performance improvements play out in real world examples with real projects, we contacted Dr. Patrick Scott of Louisiana Tech University. Dr. Scott is a professor and researcher that has been working with his senior-level Regional Economic Analysis (BUS-411) students doing real-world regional economic analysis for publication. Recently, he reached out to IMPLAN to discuss a project one of his students was working on that looked at each of the 9 MSAs in Louisiana and a combined Region for the rest of parishes (counties) in Louisiana, for a total of 10 Regions.

The project requires that the spillover effects are captured from Region to Region, which necessitates an MRIO analysis. An MRIO project with 10 Regions and many events in each Region is a large project that could require thousands of individual analyses in the previous version of IMPLAN. In fact, when Dr. Scott was initially trying to run his project, he was finding that the system couldn’t handle that amount of data processing. This made Dr. Scott’s project a perfect case study to see if the added efficiencies actually made a difference in a real-world scenario.

“IMPLAN is always relatively easy to use to implement an analysis and the newest release of IMPLAN, with the updated MRIO efficiency, is remarkably faster. This project was initially run on the old version and it was timing out [without finishing]. There were too many intermediate calculations that created a bottleneck. The latest version of IMPLAN ran the exact same model in just over three hours, which given the number of estimations, is quite an impressive gain in efficiency.”

This case study demonstrates both of the primary benefits that the new version of IMPLAN offers: (1) projects that used to be at or beyond the system’s capabilities are now processing more reliably and (2) projects are now completing noticeably and measurably faster in the new version compared to the previous version. 

If you’d like to use one of your projects as a case study yourself, take the new version of IMPLAN for a spin and let us know how these processing improvements are affecting your research by emailing us at [e-mail].

RELATED ARTICLES:

5.16 Release Notes

MRIO: Introduction to Multi-Regional Input-Output Analysis 

Written May 6, 2020