-
@MartinSGaynor @dclingi On top of more sociological explanations, I do think ABM still need to deliver. A first issue is how much "data" they produce. It's *a lot*, even for simple models. How to treat those data? On top of pure technical issues, this leads to the question of trustability.
♻️ 1 Retweets
❤️ 3 Favorites
Mood +5 🙂
-
@MartinSGaynor @dclingi ABMs are not well known, and as they produce a lot of output, how to be sure the researchers didn't pick the one simulation that prove their point, and forget everything else? More general approaches are needed. I hint toward machine learning, but it's just an intuition.Permalink
❤️ 1 Favorite
Mood -1 🙁
-
@MartinSGaynor @dclingi Another problem is a bit deeper, and is related to the question of reproducibility of computer code. One model, running in two exact machines with a small difference (like a different processor), may end up producing two sets of really different outcomes.Permalink
Mood 0
-
@MartinSGaynor @dclingi So let's say you find the code of someone's else ABM, and you want to run it again. You may find another result, and conclude that the model or the associated paper is flawed. It makes a bit difficult to differentiate good and bad models. Not good for trustability.Permalink
Mood -6 🙁
-
@MartinSGaynor @dclingi This difficulty is usually overlooked by economists, for understandable reasons: we are not computer scientists, and I've learned it through computer scientists who work on ABM. The usual lack of cooperation with other fields in econ may be a problem here.Permalink
❤️ 1 Favorite
Mood -5 🙁
-
@MartinSGaynor @dclingi That being said, I do think we now have the technology to deliver more convincing ABMs, at least in the short of middle run. But it's not done yet!Permalink
Mood 0