Indiana Jones, Economist?!

In a stunningly original paper Gojko Barjamovic, Thomas Chaney, Kerem A. Coşar, and Ali Hortaçsu use the gravity model of trade to infer the location of lost cities from Bronze age Assyria! The simplest gravity model makes predictions about trade flows based on the sizes of cities and the distances between them. More complicated models add costs based on geographic barriers. The authors have data from ancient texts on trade flows between all the cities, they know the locations of some of the cities, and they know the geography of the region. Using this data they can invert the gravity model and, triangulating from the known cities, find the lost cities that would best “fit” the model. In other words, by assuming the model is true the authors can predict where the lost cities should be located. To test the idea the authors pretend that some known cities are lost and amazingly the model is able to accurately rediscover those cities.

Here from the paper is more detail. Each step is an accomplishment and the final product is something completely unexpected. Bravo!

Our first contribution is to extract systematic information on commercial linkages between cities from ancient texts. To do so, we leverage the fact that the ancient records we study can be transcribed into the Latin alphabet, allowing all texts to be digitized and parsed. We automatically isolate, across all records, the tablets which jointly mention at least two cities. We then systematically read those texts, which requires an intimate knowledge of the cuneiform script and Old Assyrian dialect of the ancient Akkadian language that the records are written in. Taking individual source context into account, this analysis relies exclusively upon a subset of records that explicitly refer to journeys between cities and distinguishes whether the specific journey was undertaken for the purpose of moving cargo, return journeys, or journeys undertaken for other reasons (legal, private, etc.).

Our second contribution is to estimate a structural gravity model of ancient trade. We build a simple Ricardian model of trade. Further imposing that bilateral trade frictions can be summarized by a power function of geographic distance, our model makes predictions on the number of transactions between city pairs, which is observed in our data. The model can be estimated solely on bilateral trade flows and on the geographic location of at least some cities.

Our third contribution is to use our structural gravity model to estimate the geographic location of lost cities. While some cities in which the Assyrian merchants traded have been located and excavated by historians and archaeologists, other cities mentioned in the records can not be definitively associated with a place on the map and are now lost to us. Analyzing the records for descriptions of trade and routes connecting the cities and the landscapes surrounding them, historians have developed qualitative conjectures about potential locations of several of these lost cities. We propose an alternative, quantitative method based on maximizing the fit of the gravity equation. As long as we have data on trade between known and lost cities, with sufficiently many known compared to lost cities, a structural gravity model is able to estimate the likely geographic coordinates of lost cities. Our framework not only provides point estimates for the location of lost cities, but also confidence regions around those point estimates. For a majority of the lost cities, our quantitative estimates come remarkably close to the qualitative conjectures produced by historians, corroborating both such historical models and our purely quantitative method. Moreover, in some cases where historians disagree on the likely location of a lost city, our quantitative method supports the conjecture of some historians and rejects that of others.


Okay, well, they should be able to dig up lost cities if their methodology is correct.
For example, Schliemann's reading of the Iliad was considered naive by the leading German classicists, but then he went and dug up Troy.

An alternative theory to this gravity theory would be that the location of prosperous cities of the past would be fairly contingent on political and military events that would be hard to predict. I'm interested in finding out which is a better fit with the reality.

Also, I'd add that past prosperity would depend upon past geographic circumstances that might not be obvious at present. Rivers, for instance, change course and harbors fill up with silt. For example, the Greek harbors of Miletus and Ephesus in modern-day Turkey have left huge ruins behind, but they are no longer on the coast due to the rivers silting up the bays.

The location had been pointed out to him; it wasn't his idea.

Here we are from WKPD: It was Calvert who identified Hissarlik as Troy and suggested Schliemann dig there on land owned by Calvert's family.

Now I'm wondering if a gravity model could be used as a prediction tool for political and military events.

Lewis Binford began using computer modeling to do this as early as the late 1960s, it is a widespread technique in North American archaeology to find not just camp sites but also villages and other neolitihic sites, not to mention early historic sites and even 18th and 19th century military posts. It is the chief method used in Alaska. Mark Mehrer at NIU in Dekalb has written whole manuals on this stuff.

As to old world archaeology Shawn Graham was doing this with GIS and Roman itineraries in the early mid 2000s, and he has had a lot of concrete success in locating and confirming documentary evidence on the ground, not to mention predicting and finding infrastructure such as remount stations, etc... I have heard of similar work in Anatolia which was done over a few weeks and has been used for targeted surveys with considerable success.

Cruder forms of this involving basic Delbruck style analysis of travel times and march times have been used as early as the 1810s in Egypt and the Levant and even earlier in India.

What the Romans did for computer modeling:

?? .... Alex very clearly said this was a " stunningly original paper "

are you guys not easily stunned ?

1) This is pretty cool. I will retain some skepticism. I am not as impressed as Alex is with the model being able to "predict" where the cities are located given that the model was built using the data about these cities (i.e. these are in sample predictions). Don't get me wrong, if your model can't do in sample predictions well, that means your model is crap, but you also shouldn't cheer too much when it successfully predicts in sample observations.

2) While this is cool, Axtell was doing more advanced versions of this 15 years ago:

#1 I was about to write the same thing but now that I think about it being able to do in-sample predictions is probably the gold standard in economics models. I mean econ models tend to do a horrible job at predicting anything so I would expect that in economics being able to do in-sample predictions well is probably as good as it ever gets.

It's not an in sample test. It's a pseudo out of sample test. Meaning they drop the observations from the data, train or "fit" the model on the remaining data, then test the model on the dropped data.

That's not quite what they do. They are using estimates for the lost cities from the model in section 3.1 (which was built on the complete dataset) and then dropping a known city to see if they can recover its location through their estimation (a little better than pure in-sample prediction, but not much). If they were actually using the train/test paradigm (which they should have and should be utilized much, much more in economics), they would have left out some of the known cities to build the model in 3.1.

Note that they "lose" only 3 ancient cities. I would like to know what the estimates were for the cities that were not "lost".

It is amazing to see an economist be amazed that a model actually works.

So much for those pretensions of economics having anything to do with science, where a model working is considered the minimum standard, and not a source of amazement.

Can you imagine his level of glee if a climate model actually gets something right for once?

Well, considering how thoroughly poorly even the most pessimistic climate modelling regarding the Arctic has performed, I'm sure you would agree that it is quite important to spend the money to improve such models so that they at least keep up with the data reported by empirical observation.

Modeling is all well and good, but the basic terraforming is the hard part. Mankind won't permanently put an end to the ice age without a long term consistently applied effort.

'Mankind won’t permanently put an end to the ice age without a long term consistently applied effort.'

Leaving aside Dyson sphere scaled thinking, it is extremely unlikely that mankind will be able to end any of Earth's current or future ice ages 'permanently,' especially when solar output is something beyond human control.

"Gravity model" means the cities are falling? Will they ever get to the center of Earth?

Gravity model means trade is most heavy between neighbors. This is true almost always: even the famous Dutch/English East Indies spice ships did most of their trade, I've read, locally, and then once done or once in a while went back to the home country of England/Netherlands with spices, which were very profitable. But local trade was much more common.

Like the other commentators, the proof of any model is 'out-of-sample' predictions which this model does not seem to have. AlexT seems easily impressed, complete with exclamation point. Reminds me of me when I'm ignorant of something interesting, which is rare!

Bonus trivia: was Platonic the lost city of Atlantis simply a town in Thira/Santorini after the volcanic explosion of the mid-second millennium BCE? A Greek archeologist Spiros Marinatos ( ) thought so and was proved right, finding Akrotiri under volcanic rubble. He died of a heart attack right on the spot.

There is good reason to believe the fabled Atlantis and Hy Brazil islands were (as the latter name's suggests) Brazilian outposts. According to 17 th Century experts, St. Thomas visited Brazil in the late 1 th Century A.C. He may have started the trade and cultural conections between Brazil and the Old World.

@A Truth Seeker - but TR if Atlantis is mentioned by Plato (which means "dish" in Spanish) in 400 BCE, that means the ancient Greeks visited Brazil before then! Much as I admire the Greek's ability to travel the world, second only to modern Germans, I don't think that's very likely.

Actually, the Greeks visited the Egyptians from whom they learned philosophy, Math and may have heard about Brazil. Brazil was probably being visited by the famous Phoenician navigators.ávea#Archaeological_interest

Such findings change everything people thought they knew about the past. Brazil has been proved ro be an old accomplished civiliztion whose origins are lost in amidst the mists of past.

It seems unlikely that anyone would have described 4th century BC Brazil as an "advanced civilization".

There is much better reason to think that they weren't. Santorini is one plausible possibility and another is a city that once existed in Southern Portugal.

This place might have the worst comment section on the web. And that’s saying something. Look at all these awful comments.

I doubt you can do better.

It's actually one of the best. Name a better one.

What sites' comments do you normally read?

This thread excepted, the comments so far have been concise, reasonably informed, basically on topic and, by internet standards, polite.

Call me naive, but when I read something like this, I can't help thinking of fake journals:

It's positively amazing that the paper listed could have been published. Gender studies is clearly a joke. It's not remotely a science. It's a social club.

"We intended to test the hypothesis that flattery of the academic Left’s moral architecture in general, and of the moral orthodoxy in gender studies in particular, is the overwhelming determiner of publication in an academic journal in the field. That is, we sought to demonstrate that a desire for a certain moral view of the world to be validated could overcome the critical assessment required for legitimate scholarship. Particularly, we suspected that gender studies is crippled academically by an overriding almost-religious belief that maleness is the root of all evil. On the evidence, our suspicion was justified."

Yes, gender studies is ridiculous, but bear in mind that the paper was actually rejected from the top gender studies journals and was finally published in a pay-to-publish journal.

This blog's comments are the bestest ever.

Some even will read the paper. The second commenter could just throw off a mini lecture on the background of the use of this slightly obscure technique. Pretty high signal to noise ratio on this blog.

Well, that's true by blog comment standards. Not sure it's setting the bar very high. I look at the comments here in equal parts for insight and amusement.

Well, trade seems to work almost like according to rules of physics, Ricardo’s law of comparative advantage is so evident when learned and so mind-opening... and still people vote for Trump and other assorted protectionists. Of course statist schools and curricula are to blame, but it is difficult to forget H L Mencken aphorism about democracy being the art and science of managing the circus from the cage of the monkeys.

I will be impressed when they actually manage to find a previously unknown city using this model.

I come from natural sciences background and when reading this blog, I'm often dumbfounded by how low the standards for evidence are in economics. Being able to predict something that was used to construct the model is not yet a victory; at best it's an indication that there is nothing obviously and horribly wrong with the model. Technical ingenuity or novelty also do not make a model good. The only thing that matters in the end is how well the model is able to make predictions that correspond to empirical findings. Findings that were known to the people who constructed the model should count for less than completely new findings because it is always possible to construct an overfitted model that predicts everything that is already known perfectly but produces nonsense otherwise.

The paper indeed presents an interesting idea, but I would really like to see an actual lost city located using this model and dug out from the ground before getting too excited.

Comments for this post are closed