00:00 - 00:03 | We are now ready to present our identification strategy |
00:04 - 00:05 | We used a natural experiment |
00:05 - 00:07 | from the fall of the Berlin Wall |
00:08 - 00:12 | to obtain exogenous variation in access to markets |
00:12 - 00:15 | for both new and used West-German automobiles. |
00:17 - 00:19 | I assume the model |
00:19 - 00:21 | is estimated using NFXP |
00:24 - 00:26 | My Führer... |
00:27 - 00:28 | Stata... |
00:31 - 00:33 | Stata does not have NFXP |
00:34 - 00:36 | We used standard diffs in diffs |
00:53 - 00:58 | If you agree inference without theory is blind. Please leave the room |
01:13 - 01:15 | Are you completely out of your mind? |
01:15 - 01:17 | Have you never heard of Harold Zurcher? |
01:18 - 01:23 | Even a student can estimate this simple model |
01:25 - 01:28 | It is a simple combination of the BHHH algorithm |
01:29 - 01:31 | and Newton-Kantorovich iterations |
01:31 - 01:34 | You just need to use the implicit function theorem |
01:34 - 01:37 | and the Frechet derivative of the Bellman operator |
01:37 - 01:40 | is so simple to compute for this model. |
01:40 - 01:42 | But we have to submit our paper to the QJE |
01:42 - 01:46 | Even QJE can see that this is completely useless econometrics. |
01:46 - 01:48 | My Führer, I'm soon up for tenure. |
01:48 - 01:52 | Econometric work requires hard work |
01:53 - 01:54 | You need economic theory! |
01:56 - 01:57 | Go study some econometrics |
01:57 - 02:00 | and try to understand data and institutions |
02:00 - 02:03 | and learn numerical methods and computer programming |
02:04 - 02:08 | Then let's talk about whether your so-called identification strategy |
02:08 - 02:13 | can tell us anything about the German automobile market |
02:14 - 02:16 | A car is a durable asset. |
02:17 - 02:21 | What did you think off. Stalin knew how to produce cars. |
02:27 - 02:29 | How can you ignore that? |
02:30 - 02:34 | Consumers are heterogeneous, forward looking and trade with each other |
02:34 - 02:36 | This is a dynamic equilibrium. |
02:41 - 02:42 | SUTVA does not hold. |
02:43 - 02:47 | How could you even consider any treatment effects approach |
02:48 - 02:53 | You need to start coding the dynamic program now! |
02:54 - 02:56 | This problem is easily done in MATLAB. |
02:56 - 02:59 | No fancy GPU computing, OpenMP or HPC of any kind. |
03:00 - 03:02 | I can code this while driving my Volvo. |
03:04 - 03:07 | Don't worry, BLP is OK |
03:14 - 03:16 | Start with Harold Zurcher |
03:19 - 03:23 | Just a few car types and a finite number of vintages |
03:25 - 03:26 | Add transactions costs |
03:31 - 03:33 | Allow for driving and fuel taxation |
03:40 - 03:46 | Then consider flexible specifications of car characteristics |
03:46 - 03:49 | Add some more heterogeneity and impose equilibrium |
03:53 - 03:56 | Estimate using NFXP and run counterfactuals |