"Why aren't there more simulations in studying or understanding things that impact the society?" [especially economics]
Firstly, because economists' beliefs are wrong or not-even-wrong; second because the discipline of engineering simulation software is somewhere between incompetent and nonexistent.
Your post is important, I'm glad you restacked it recently, but it tries to deal with too much to comment on briefly. I'll just give some bullet points:
* Agent models have to have populations of agents which pay attention almost only to their own immediate economic concerns, not to aggregate statistics. The aggregates have to emerge from the decisions of the agents.
* The economy as a whole is the wrong place to start. Try something simpler, such as people trading a given currency pair, such as EUR / USD.
First, though, try hand-trading a practice account (e.g. Oanda using Metatrader 3 software). Give yourself 100,000 times more trading experience and discipline using automated algo-generation and testing software that lets you try all sorts of variations, for instance ForexStrategyBuilder (an old c. 2008-12 demo version is best. The web version has fatal drawbacks, and FSB Pro is expensive.).
You'll find this simplest of all markets still isn't simple at all, but the strategies that have a chance to work out-of-sample ARE very simple. (e.g. the ur-mean-reversion strategy: on a ranging (flat) 1 or 5 minute candle chart, sell the minimum broker-allowed amount each time the price crosses an upper "steady" indicator band, buy a minimum amount each time crossing the lower band, never close, no stops. Set the band spacing so that it trades between every 10-100 candles, it will automatically scale in and out of positions over a period of days or weeks.)
Model the effects on agent and market behavior of differing market rules in different jurisdictions, e.g different margin limits, actual interest rates, trading provider behaviors, regulators etc. Model the effects of different market participants using different strategies, including both psychologically-affected hand-trading and various algorithmic approaches. After tuning your simulation so it bears a decent resemblance to high-time-resolution market price- and depth-of-book data over modest intervals when there are no news shocks, expand the simulation to include effects from quantitative news events (economic numbers announcements and some other correlated major currency pairs, then other financial time series. This is still just a warm-up for simulating an economy.
* You'll want to get each of your agents to fit in a GPU core or at least a small block of cores to be able to run enough of them often enough. The time-steps have to be minutes. or at most an hour for a realistic economic simulation, daily won't do.
* Agents in the population can't all be alike, or even similar, whether in specific private data, types of data (some may not pay attention to most prices of things they have or want, others may watch every tick of many prices), models of their world (and most may have incorrect fundamental models, e.g. believing that banks loan out deposits), ways of making decisions (e.g. emotional, stochastic or algorithmic); ability to make correct decisions or, equivalently, the difficulty of problems which they can solve. (see my posts on Rasch measures of intelligence).
* Once you have such an agent-based simulation of a small number of financial time-series working to any useful degree, you'll be making too much money with it to even think of publishing it. This is why the literature of quantitative finance is usually of such low quality. The legendary Ed Thorp discovered the options pricing formula years before Black and Scholes, almost published, but didn't. (Any readers who don't know about him - it's worth looking Ed Thorp up. Math prof, invented the true market-neutral hedge fund, ~invented counting cards, invented the wearable computer with Claude Shannon to beat roulette, etc.)
"Why aren't there more simulations in studying or understanding things that impact the society?" [especially economics]
Firstly, because economists' beliefs are wrong or not-even-wrong; second because the discipline of engineering simulation software is somewhere between incompetent and nonexistent.
Your post is important, I'm glad you restacked it recently, but it tries to deal with too much to comment on briefly. I'll just give some bullet points:
* Agent models have to have populations of agents which pay attention almost only to their own immediate economic concerns, not to aggregate statistics. The aggregates have to emerge from the decisions of the agents.
* The economy as a whole is the wrong place to start. Try something simpler, such as people trading a given currency pair, such as EUR / USD.
First, though, try hand-trading a practice account (e.g. Oanda using Metatrader 3 software). Give yourself 100,000 times more trading experience and discipline using automated algo-generation and testing software that lets you try all sorts of variations, for instance ForexStrategyBuilder (an old c. 2008-12 demo version is best. The web version has fatal drawbacks, and FSB Pro is expensive.).
You'll find this simplest of all markets still isn't simple at all, but the strategies that have a chance to work out-of-sample ARE very simple. (e.g. the ur-mean-reversion strategy: on a ranging (flat) 1 or 5 minute candle chart, sell the minimum broker-allowed amount each time the price crosses an upper "steady" indicator band, buy a minimum amount each time crossing the lower band, never close, no stops. Set the band spacing so that it trades between every 10-100 candles, it will automatically scale in and out of positions over a period of days or weeks.)
Model the effects on agent and market behavior of differing market rules in different jurisdictions, e.g different margin limits, actual interest rates, trading provider behaviors, regulators etc. Model the effects of different market participants using different strategies, including both psychologically-affected hand-trading and various algorithmic approaches. After tuning your simulation so it bears a decent resemblance to high-time-resolution market price- and depth-of-book data over modest intervals when there are no news shocks, expand the simulation to include effects from quantitative news events (economic numbers announcements and some other correlated major currency pairs, then other financial time series. This is still just a warm-up for simulating an economy.
* You'll want to get each of your agents to fit in a GPU core or at least a small block of cores to be able to run enough of them often enough. The time-steps have to be minutes. or at most an hour for a realistic economic simulation, daily won't do.
* Agents in the population can't all be alike, or even similar, whether in specific private data, types of data (some may not pay attention to most prices of things they have or want, others may watch every tick of many prices), models of their world (and most may have incorrect fundamental models, e.g. believing that banks loan out deposits), ways of making decisions (e.g. emotional, stochastic or algorithmic); ability to make correct decisions or, equivalently, the difficulty of problems which they can solve. (see my posts on Rasch measures of intelligence).
* Once you have such an agent-based simulation of a small number of financial time-series working to any useful degree, you'll be making too much money with it to even think of publishing it. This is why the literature of quantitative finance is usually of such low quality. The legendary Ed Thorp discovered the options pricing formula years before Black and Scholes, almost published, but didn't. (Any readers who don't know about him - it's worth looking Ed Thorp up. Math prof, invented the true market-neutral hedge fund, ~invented counting cards, invented the wearable computer with Claude Shannon to beat roulette, etc.)