Complexity and decline

[Comrade Ceasefire]

What the world needs now (apart from fewer folk singers,) is a form guide for economists, similar to what is available for horse races.

If economists make predictions, the public should be able to see how their prognostications fared, and whether they are any better than Mystic Meg's current replacement, the toss of a coin, or Raven .

Most successful chimpanzee on Wall Street

www.guinnessworldrecords.com

 

[Samson]

Keep in mind, Samson, economics is one of the social sciences. Those can never be nearly as exact as the physical sciences because of what Voidwalkin said: people are irrational. At least they don't move as predictably as water does.

A) So much of medicine is to do with the irrationality of humans.

B) It does not matter to the stats, it just means your confidence interval is bigger.

I shall try and give an example I was looking at before I go to bed. The current UK estimates/prediction of GDP is below, for 2019 - 2027 given to 4 significant figures. I suspect these exist for every quarter for say 2 decades. Given just that data, some fairly non-informative assumptions about the error term and priors and some Bayesian integration make an estimate of the confidence interval of the predictions.

Obviously if one had access to the models with which these numbers were generated one should be able to come up with a much better estimate. Why is this information not available?

Spoiler GDP :

DATE,GDP 1,GDP2
2019Q1,555.1,555.1
2019Q2,556.9,556.9
2019Q3,561.0,561.0
2019Q4,560.9,560.9
2020Q1,545.6,545.6
2020Q2,434.7,434.7
2020Q3,507.6,507.6
2020Q4,514.5,514.5
2021Q1,509.3,509.3
2021Q2,546.6,546.6
2021Q3,556.0,556.0
2021Q4,564.4,564.4
2022Q1,567.4,567.4
2022Q2,567.9,567.9
2022Q3,567.4,567.4
2022Q4,568.0,568.0
2023Q1,569.0,569.0
2023Q2,569.1,569.1
2023Q3,568.4,568.4
2023Q4,566.6,566.6
2024Q1,568.6,568.6
2024Q2,570.0,569.3
2024Q3,571.3,569.5
2024Q4,572.6,569.6
2025Q1,573.9,569.7
2025Q2,575.3,570.0
2025Q3,576.7,570.5
2025Q4,578.3,571.3
2026Q1,580.0,572.3
2026Q2,582.0,573.7
2026Q3,584.2,575.4
2026Q4,586.5,577.3
2027Q1,589.0,579.4
2027Q2,591.5,581.7

"GDP 1, £ billion chained-volume measure (reference year 2019)",,
,,
,"GDP 2 GDP projections based on market interest rate expectations, other policy measures as announced","GDP projections based on interest rates constant at 5.25%, other policy measures as announced"
,,

 

[Zardnaar]

I believe Samson's point r.e. rigour doesn't rely on strict equivalence between social and physical science.

Or at the very least, if we accept this wiggle room, then tying cause and effect in concrete terms is, at best, subject to the same wiggle room and therefore the correlation (r.e. UBI and inflation) is probably not as absolute as it was phrased.

What we are saying is UBI is most likely inflationary. So are tax cuts.

UBI on a massive scale it's debatable if it will work.

We know cheap housing works.

So it's opportunity cost. Is UBI worth it vs a welfare state with say an effective social housing build program.

And we can't really afford both. UBI even a moderate one eats up all the other social programs. And in effect its still a subsidy. 30 odd years later they don't work so well (government should have done xyz instead).

It was only cheaper in the short term (money saved while benefitting from existing social stock).

Yangs UBI was a low one, cost a bomb and was less than what we pay people on welfare . So it basically makes things worse for the poor.

A more generous UBI is unaffordable even with massive tax hikes.

If housing is cheap it still benefits the middle class either via cheap rent or mortgages.

 

[Voidwalkin]

A) So much of medicine is to do with the irrationality of humans.

Which reminds me; when empiricism is asked to measure what is beyond its instruments to observe and quantify(emotion) it faces a reproducibility crisis(psychology and the social sciences)

B) It does not matter to the stats, it just means your confidence interval is bigger.

I suspect that pre-Brexit modelers analyzing the future would have had to calculate things like future immigration rates, willingness to invest in a new economic situation.

Which effectively involves taking a snapshot of public opinion not just at a point in time, but over decades; this is next to impossible. There will be large deviations from past patterns that snowball away from the model as time goes on.

And of course Brexit itself would have likely been unforeseen by modelers 5 years prior to.

 

[innonimatu]

Economists... shamans liked telling people the world was ruled by complex occult forces, so they had to heed the shaman's gudance.

Much of modern complexity is the same think with new clothing. Technically neddlessly but oh so very useful to its proponents.

 

[Farm Boy]

Keep in mind, Samson, economics is one of the social sciences. Those can never be nearly as exact as the physical sciences because of what Voidwalkin said: people are irrational. At least they don't move as predictably as water does.

It's more like herding.

 

[Zardnaar]

Economists... shamans liked telling people the world was ruled by complex occult forces, so they had to heed the shaman's gudance.

Much of modern complexity is the same think with new clothing. Technically neddlessly but oh so very useful to its proponents.

Sorta.

Roman's had inflation as well along with one of the Chinese dynasties from printing to much money.

Populist like to ignore it but supply and demand is very real.

Lots of political and economic systems have struggled with it. Roman's debased their coins, Chinese over printed money. Spanish dug up to much gold and silver in the new world. Mansa Muta spent to much gold. Soviets you saved money because you had consumer goods and food shortages.

 

[Gori the Grey]

It's more like herding.

Yeah, and show me the mathematical model for getting a group of animals from one point to another.

 

[Samson]

Yeah, and show me the mathematical model for getting a group of animals from one point to another.

In 5 mins:

Controlling noisy herds
The wisdom of the crowd breaks down in small groups. While large flocks exhibit swarm intelligence to evade predators, small groups display erratic behavior, oscillating between unity and discord. We investigate these dynamics using small groups of sheep controlled by shepherd dogs in century-old sheepdog trials, proposing a two-parameter stochastic dynamic framework. Our model employs pressure (stimulus intensity) and lightness (response isotropy) to simulate herding and shedding behaviors. Light sheep rapidly achieve a stable herding state, while heavy sheep exhibit intermittent herding and orthogonal alignment to the dog. High response isotropy enhances group cohesion but complicates group splitting. We construct a unified phase diagram for sheep behavior, identifying three regimes (fleeing, flocking, and grazing) based on group size and stimulus specificity. Increasing stimulus specificity shifts small group behavior from grazing to fleeing, while larger groups exhibit flocking. This transition underscores the challenge of controlling small indecisive collectives. Introducing the Indecisive Collective Algorithm (ICA), we show that deliberate indecisiveness and stochasticity improve control efficiency. ICA outperforms traditional averaging-based algorithms in high-noise settings and excels in tasks requiring group splitting. Our study offers a foundational framework for controlling small, indecisive groups, applicable to biochemical reactions, cell populations, and opinion dynamics.

Spoiler Dynamics of Sheep Herding and Shedding :

Spoiler Legend :

I like that paper, but this may be more exactly what you asked for, though it does not have real life validation:

Spoiler Space mapping-based receding horizon control for stochastic interacting particle systems: dogs herding sheep :

 

[Farm Boy]

Well, I read what your posted, but you're going to need to sell me they aren't just playing with dreams of electric sheep.

 

[Gori the Grey]

In 5 mins:

That's great except that it involved no, you know, sheep. It does reveal clearly how a computer model of sheep operates.

You will not get your numbers around my will, your addition around my volition.

 

[Hygro]

I wonder: will there come a point where systems in place become complex beyond our comprehension?

Are we already there?

Take economics. I'm not an economic genius. But you can tell when people talk economics that their grasp is variable. Some evaluate the board like a chess grandmaster, some are hopelessly flawed.

Related: also concerning to me is that the sum of knowledge required to speak intelligently about some complex topics necessitates such dedication that expertise is a prereq. Expertise can lead to a trap.

Example. If the question is asked, what are the dangers of AI, you'd need to bring in an AI expert to understand what AI is/will likely be capable of. Then a different expert to understand how companies are likely to adapt the technology, given cost barriers, expected RoI. Then, if necessary, a politician to even have a sense of which(if any) legislative policies can gather enough support to pass.

Terrifyingly, all these experts must be right in their assessments and acting in good faith, or society will likely be paralyzed to inaction, act w/o effectiveness, or act counterproductively. You might be better off turning to generalists if errors become more likely as experts interact with each other.

I can totally see a significant stressor coming along a weak link in the chain breaking, setting in motion disastrous turns of events.

I agree that domain knowledge and summed domain knowledge is difficult and itself very interesting but I don’t agree the system is so fragile and unable to aggregate the divergent hunches of amateurs to the point of systemic expertise that we are in any danger.

 

[Comrade Ceasefire]

That's great except that it involved no, you know, sheep. It does reveal clearly how a computer model of sheep operates.

You will not get your numbers around my will, your addition around my volition.

Getting every individual sheep's position correct over any significant time period is impossible.

OTOH, if there were many more sheep, of the order of the number of molecules of gas in a container, there's no need to get the individual positions correct for a meaningful, useful answer.

 

[Birdjaguar]

Yeah, and show me the mathematical model for getting a group of animals from one point to another.

This is the best I can come up with.

 

[Gori the Grey]

So I'm going to propose a formula. It's an inverse ratio.

In any given study, the maths can be robust to the extent that the study does not concern volition.

Water "seeks its level," but human beings seek food . . . shelter . . . companionship . . . distraction . . . purpose . . .

The one you can get your numerical lasso around. The other . . . not so much.

 

[Tee Kay]

all i'm getting from this thread is that we should replace economists with border collies

 

[Farm Boy]

Getting every individual sheep's position correct over any significant time period is impossible.

OTOH, if there were many more sheep, of the order of the number of molecules of gas in a container, there's no need to get the individual positions correct for a meaningful, useful answer.

This is the answer of a bad, bad, shepherd.

This is the best I can come up with.

View attachment 694501

Just breed an entire species of hypersocial and cooperative pack animals that ride the confluence of smart and motivated! Then sic a pack of them on the task each time.

It's kinda remarkable, really.

 

[PhroX]

Economics is hard because you need a good model grounded in economic theory, a good way to solve that model, good data to calibrate parameters of the model, and to have chosen the correct set of things to include in your model to get significant results. Almost no paper can get all of these perfectly right but that doesn't mean all papers are worthless. "All models are wrong but some are useful"

Why do you need to have your model grounded in theory? Get a large enough dataset, then use that to build an empirical model. It doen't matter what theory it's based on as long as it works.

Also, I'm not really looking for a paper describing how it works. I'm looking for commercial modelling software that does it. If I was showing someone how computational hydraulics is done, I wouldn't find something a univerisity professor had made, I'd point you at tools like ANSYS CFX, InfoWorks or FloodModeller (formely, unfortunately, known as ISIS). It just really surprises me that this kind of thing doesn't exist given the amount of data we have on economics - I would've thought that some predictive modelling programmes based on this empirical data would exist.

As an aside, this is one of the areas where AI is beginning to have some useful results - making predictive models from huge datasets.

 

[Samson]

Well, I read what your posted, but you're going to need to sell me they aren't just playing with dreams of electric sheep.

That's great except that it involved no, you know, sheep. It does reveal clearly how a computer model of sheep operates.

You see that they built their model and then tested it against data from real competitive sheep herding, comparing the method against another one and the results are objectively better.

Spoiler Validation :

 

[Farm Boy]

...and?