Serving the Quantitative Finance Community

 
User avatar
katastrofa
Posts: 7475
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Re: Is economics a science?

January 18th, 2024, 10:57 pm

I can think of two situations taking place when you feed highly correlated features to a NN model.
1) It may recognise that one variable is redundant and assign minimal weights to it - and you can usually see it in the first layer. Additionally, e.g. if you add an attention layer to the model, you can see that the model ignores some of the features and is "streamlining" to the most informative ones. It's a sign that you should revisit the feature selection stage - to improve the training efficiency and avoid overfitting.
2) Related to your point 3, the multi-layer structure and non-linear activation functions enable NNs to build abstract representation of the data you feed to them. They can learn more intricate relationships, like X*Y^2 for Y < 121 and X^2*Y^3 for Y>=123 or whatever, that a linear model obviously can't capture.
Now that we have a set of methods to look at the model decisions, you can see what exactly is happening within the NN and even try to explain the impact of individual features - or their interplay (I attach the slide with a list of such most popular methods, which you have probably googled yourself already!).
Vienna2023_slide9.png
Having said that, while multicollinearities aren't generally a problem for prediction purposes, from my experience, if the goal is interpretation or you have reasons to believe multicollinearities are an issue, one should handle them before training, like in the case of linear models. NN approach is better but still case specific, as you said. Essentially, there's no free lunch!
 
leptoq
Posts: 366
Joined: February 15th, 2018, 4:42 pm

Re: Is economics a science?

January 19th, 2024, 8:54 am

Thank you for sharing, katastrofa! This is very useful! 
I'm using actively at this point: SHAP, Counterfactual and some Model-Agnostic Methods, but certainly should fill gaps in my education about other approaches that you mention above. 
 
User avatar
DavidJN
Posts: 244
Joined: July 14th, 2002, 3:00 am

Re: Is economics a science?

April 6th, 2024, 7:34 am

A graduate program in economics I attended long ago began with a month-long, six-hour a day math/stats review course led by a genius and real gentleman named Larry Epstein. At the end of the course, the learned professor said “If I want you to leave with one thing after the month we’ve just spent together, it is that this here ain’t science boys and girls, this here is religion!”
 
What I take to be your main and intriguing point is that science is immutable in the sense that belief cannot change things per se, whereas belief and confidence have a great deal to do with the outcomes of markets and economics. Belief isn’t science, it seems. How can we square this with the notion in physics that the simple act of observing physical processes can change how they behave? The so called the Heisenberg Principle. Isn’t that a direct analogue in science of your main complaint about economics and markets? That something people do, in this case mere observation, can change how elemental things behave.
 
Is science as immutable and as successful as you suggest? We surely call physics a science, but it too has lots of issues. A big one would be quantum mechanics and gravity. The Standard Model of Physics reads like Darth Vader orating in Star Wars… “Luke… there is a Strong Force and a Weak Force!”. Add the puzzles re. dark matter and dark energy to the list. I’ve long wondered if physics has stalled and is even regressing by focusing too much on the small, and I am glad to see recent work extending Newtonian physics toward a complete theory by incorporating selected randomness.
 
It seems to be not merely the existence of belief and confidence in markets and economics that concerns you, but also that the market beliefs are frequently wrong. Paul Samuelson famously said that economists successfully forecasted nine out of the last five recessions. But that depends on what you mean by the word wrong.
 
Are markets and economics frequently wrong or simply changing in response to the arrival of new information? Are you equating volatility in the underlying drivers of economic activity with forecasting errors? Systemic forecasting errors lead to arbitrage money machines that are actually quite rarely found in practice. Frequently revised beliefs reflected in volatile market outcomes are not in themselves evidence of logical weakness, they might simply mean that the environment in which economics operates is dynamic and very volatile. Having said this, the empirical fact of the occasional so-called flash crashes in the stock market does support your fundamental view.
 
Now some thoughts in favour of your position.
 
The traditional path to developing confidence in the outcomes of markets and economics, in turn justifying the application of scientific methods to the field, is competition, which is an appeal to the law of large numbers. There is considerable weakness in this belief. The formal “proof” that the competitive markets foundation of capitalism leads to optimal resource allocation is a ridiculous textbook fiction. It is truly ridiculous, as it assumes all the real problems away. The subfield behind it is welfare economics. It is also predicated on a strict assumption of Pareto optimality,  in which winners compensate losers. In what world does that happen? Certainly not this one.
 
Further, the competitive market structure that leads to confidence in markets and price discovery is increasingly a fiction as wealth inequality widens. A surprisingly small number of people own a surprisingly large amount of the stocks. That is not meaningful competition. Every morning on a trading desk I worked at, someone would cry out at the stock market opening bell “The casino is open!” and similarly cry out “The casino is closed!” at market close. We genuinely meant it. The stock market is increasingly a joke. My personal belief is that it is fundamentally overvalued (propped up by lack of competition), but as Keynes famously said, market irrationality may persist longer than you may remain solvent betting against it.
 
One can make a strong argument that increasing direct government interference in markets – (e.g. extended period of imposed artificially low interest rates and highly prescriptive and expensive banking capital and liquidity rules) has distorted market outcomes away from their competitive ideal.
 
Re. your comment on the Age of Enlightenment - have you read John Ralston Saul’s Voltaire’s Bastards The Dictatorship of Reason in the West? He puts down our modern management problems (note it was written in the 90‘s) to specialization of the knowledge released by the Enlightenment, and the corporate form of business organization rather than political pollution per se. Just an alternate view for you to consider.
 
A final observation, a nit really, is that English is probably not your first language. You are clearly conveying your ideas, but you’ll want someone to polish the grammar somewhere along the way.
 
 
User avatar
katastrofa
Posts: 7475
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Re: Is economics a science?

April 6th, 2024, 4:58 pm

“ science is immutable in the sense that belief cannot change things per se, whereas belief and confidence have a great deal to do with the outcomes of markets and economics”

Einstein to Heisenberg on his early results on quantum mechanics interpretation: “It is the theory which decides what we can observe. You must appreciate that observation is a very complicated process. Only the theory that is knowledge of natural laws enables us to deduce the underlying phenomena from our sense impressions”.

Science is pretty mutable, at least according to Einstein or the Duhem-Quine hypothesis, to which he actually referred. The Nature isn’t. But then again, think of collapse/ observer effect or how the constructivist epistemology puts it. We know nothing, sigh.