> [Ray Dalio] dispenses with the way economists have long taught economics in school, and instead explains the economy as if it were a “machine” that he believes is much easier to understand and predict.
"Dispenses", really? Dalio is no small force [1], but his thoughts and models are, in some sense, just "another" model to add to the mix.
[1]: http://www.economist.com/node/21549968
Here, my point is not to support or detract from Dalio's model. My point is simply that many people (and many journalists) don't seem to get what models are, at their core. They are tools.
Yes, sometimes traditional economic models get "too much" credit and mindshare. I like seeing alternative models. After taking a look at Dalio's snappy presentation, connecting almost everything with a nice explanation, I really miss an academic presentation, with equations. (I'm not saying Dalio doesn't have them -- Bridgewater Associates certainly does.)
Both in the case of this presentation and in many mainstream accounts, the language commonly used in economic models sometimes lulls readers/listeners/viewers into conflating the model with reality. Don't let it. The world is complicated. If you want to force a complicated world into a simple model, you can, with varying results.
Here is my central point. Everyone, including all flavors of economists and Dalio himself, are peddling models. Don't let their claims of being "simple" and "mechanical" distract you. (Some people claim the opposite, e.g. "my complex model is the most realistic".) People will accentuate any aspect of their model. Like any good salesperson or marketer, they will find the words that engender trust.
(In modeling-speak, modelers often seek to build models that their audience will find intuitive. Models that are too non-intuitive, in their assumptions at least, sometimes get quickly discarded. Models with intuitive low-level behavior and non-intuitive higher-level behavior often garner a lot of attention, because they are deemed to be believable but surprising.)
What makes a good model? That's a long conversation. For now, I'll just say this: use some model(s) that are useful for your situation. It is obvious, but you'd be amazed how few people take this advice to heart. Perhaps they want the "best" model, which is akin to asking for the "best" car. In practice, many people go with the most familiar model, though many don't like to admit it.
I'm not saying models are bad. I am saying two main things:
1. If you are using numbers (even rough ones) and seriously think that you aren't using a model, you are fooling yourself. I'll give three examples where people pretend that they aren't using a model:
1A. If you are shooting from the hip, then you are using an unspecified, perhaps instinctual, probably non-repeatable, model.
1B. If you are "just doing the numbers" then your choice of what numbers to include is your model.
1C. If you are doing "theory-free" data mining, then you are still relying on some technique(s) to surface certain patterns. Ok, so you might not be using a specified model (such as linear regression with particular variables), but how did you select your variables? If you are using a SVM, what kernel are you using? If you are using a NN, how many layers and what configuration? These are still assumptions. Your choice of technique (and preference for bias vs. variance) will shape what patterns you find.
2. Relying on too few models is foolish.