jrissman
commented
3 years ago I'm upping this one to "high priority" because I think it is important and structurally tractable.
Let me know if you are able to provide a time-series version of io-model/BObIC (the Excel file). It should not just scale every ISIC code up or down in proportion to future GDP changes, but rather, it should show which ISIC codes are likely to grow or shrink over time in the BAU case. For instance, coal mining should get smaller over time even though total GDP increases. If you can supply that Excel file, I think I can adjust all the rest of the IO variables structurally within Vensim to enable time-series performance across all IO variables, even the DLIM/TLIM matrices.
robbieorvis
Several data sources have output by ISIC code in future years, including for the US and EU models. We discussed adding in the structure to correctly handle this data. Full discussion below:
Hi Jeff,
Thanks for this and thinking through this. You definitely bring up a lot of things I hadn’t considered!
Some of the datasets may have the value added component, but they almost certainly don’t have the other components you mention, so we would have to scale.
I also realized that to complicate things further, we have projected labor productivity rates for each sector as well, and that would also have to be factored in, meaning it may not be as simple as scaling value added and holding the percent of labor constant (we could rescale the percent of labor, but the question is then what do the industries do with the reduced value add spending? Do they drop prices? Do they hold prices constant but increase profit? Etc…). It’s definitely tricky!
I would be open to thinking through it some more though. Perhaps we can spend some of our check-in tomorrow thinking about how this might work.
There are a couple of reasons I brought this topic up (in addition to improving accuracy): in the EMF working group, one of the sets of outputs they are asking for is projected future output by industry, so that leads me to believe that many models have this data, though my guess is that they have it to estimate energy demand, with an energy intensity multiplier, and not that they are trying to estimate employment; another is that I have been asked specifics about outputs from different industries in future years but currently don’t feel confident answer that because we are really only estimating the change from some future year and not the total output since we are holding today’s output constant.
Anyway, looking forward to discussing more!
-R
Robbie Orvis Director of Energy Policy Design Phone: 415-799-2171 98 Battery Street, Suite 202 San Francisco, CA 94111 www.energyinnovation.org
Check out our book, Designing Climate Solutions: A Policy Guide for Low-Carbon Energy Available wherever books are sold
From: Jeffrey Rissman jeff@energyinnovation.org Sent: Wednesday, May 19, 2021 1:09 PM To: Robbie Orvis robbie@energyinnovation.org Subject: RE: Should BAU Output by ISIC Code be time series?
Hi Robbie,
Do the BAU data you’re talking about capture shifts in the percentage of output from different ISIC codes? For instance, do they show ISIC 05 (coal extraction) having a smaller share of total output in future years than ISIC 05 has today?
Do you have similar data for the other variables we use along with output? For instance, BAU Value Added, BAU Employment, and BAU Employee Compensation? These values would get out-of-sync if you update BAU Output but fail to update those others. For instance, it would show that coal output drops, but somehow coal’s value added, employment, and employee compensation remain the same.
If you don’t have data for those three, we could scale them all to match the changes in output in percentage terms. But that’s harder to do with DLIM and TLIM. For instance, suppose your data show ISIC 20 (chemicals) grows in future years. We can assume employment, value added, and employee compensation grow proportionately. But for DLIM and TLIM, who buys those new chemicals? We could assume it’s all the same buyers as we have today, but scaled up proportionately, but that wouldn’t do. I think the thing to do would be to first scale the columns (the buyer industries) according to changes in BObIC, and then scale the rows (the seller industries) according to changes in BObIC. We’d have to see if it works. (I can do the scaling inside Vensim so that we don’t have to split DLIM and TLIM into dozens of different CSV files. This would be much cleaner than trying to do it in Excel. For that matter, I could similarly do the scaling of BAU jobs, value added, and EE compensation in Vensim, rather than requiring those variables to be supplied as time-series, minimizing the increase in difficulty of obtaining input data across the various EPS regions.)
I do think this is doable if you have BAU data on output by ISIC code that shows the output of each ISIC Code in each future year (and is intelligent about it – i.e. it’s not just blindly scaling up all industries’ outputs by the overall change in national GDP). It is worth doing.
But you should be aware this is not as quick an update as your message seems to imply, because it requires the whole suite of I/O variables to be updated together, so they remain in-sync. Just updating output without touching the others would get things out-of-sync and break stuff. But I think the others can be estimated inside Vensim rather than scaled in Excel if you don’t have explicit projections, and this might be better anyway, because it ensures that they remain in-sync (whereas using different input data files means there is a risk that the different data sources don’t line up, and if that happens, you can get runaway behavior in the energy use feedback loops that is very difficult to debug).
Jeff
From: Robbie Orvis robbie@energyinnovation.org Sent: Wednesday, May 19, 2021 6:19 AM To: Jeffrey Rissman jeff@energyinnovation.org Subject: Should BAU Output by ISIC Code be time series?
Hey Jeff,
Quick question for you. I noticed that BAU Output by ISIC Code is a time invariant input file, but I am wondering if it should be a time series file? With the new data we are using in the US from EIA, we have output as a time series by industry. Other data sets, like the one we are using for the EU, also have this as a time series. I think that the decision to make it time invariant was that the OECD data is for a single year, but we could also format the file as a time series and use static data for regions without projections. What do you think?