Skip to Simulator →
Scroll to begin
← → or scroll
Section I
2008
The year that changed central banking
$0 trillion
Global wealth destroyed
World Bank, 2009
0 million
Jobs lost worldwide
ILO, 2009
0 million
Families lost their homes
CoreLogic, 2008–2012
0
US banks failed
FDIC, 2008–2010
What the world's best economic models predicted
IMF World Economic Outlook — April 2008
+3.8%
Global GDP growth predicted for 2009
"Risks have increased, but growth remains solid"
VS
Actual 2009 Global GDP
−0.1%
The worst contraction since World War II
IMF WEO Table 1.1 — 3.9 percentage point error
It kept happening
Greece — IMF Bailout, 2010
+1.1% → −7.3%
GDP prediction vs reality for 2012
8.4pp error. IMF later admitted multipliers were 3x too low
COVID — IMF Outlook, 2019
+3.4% → −3.1%
Global GDP prediction vs reality for 2020
No model predicted COVID — but that's the point. IMF WEO Oct 2019
Inflation — Federal Reserve, 2021
2.4% → 5.8%
PCE inflation prediction vs reality
Powell retired the word "transitory" on Nov 30, 2021
2008. 2010. 2020. 2021. — The same framework failed every time.
WHY?

Because every major central bank was using the same type of model.

It's called DSGE — Dynamic Stochastic General Equilibrium.
The standard tool of every central bank in the world.

Flaw 1 of 4
→
1
Millions of people become one
DSGE compresses an entire country — unique households, banks, traders, farmers — into one "representative agent" with average income and average behavior.
What this means

A billionaire and a street vendor become one consumer with "average" wealth. A bank run becomes impossible — the agent can't withdraw money from itself.

Flaw 2 of 4
What DSGE assumes
f(x)
Every person solves complex optimization equations and has perfect knowledge of the economy's future
VS
What people actually do
$?!
People panic-buy gold when they hear rumors. Neighbors copy neighbors. Crowds follow crowds.
Assumes perfect rationality
Real people don't optimize — they panic, follow neighbors, trust rumors, and make decisions based on emotion, habit, and fear.
Flaw 3 of 4
BANK
×
No banks. No contagion. No crises.
The canonical DSGE model had no financial sector — no banks, no credit, no leverage. With one agent, panic can't spread and the "E" in DSGE means the model assumes the economy always returns to equilibrium.
In other words

A weather forecast built on the assumption that storms cannot happen — using a model that doesn't include clouds.

Flaw 4 of 4
How the model is built
~
Calibrated on 20-30 years of stable, calm economic data
VS
What it's asked to predict
!
The one extreme event that breaks every pattern in that data
Trained in calm waters, tested in storms
DSGE uses linear math — straight-line approximations that work for small fluctuations. But crises are massive, non-linear events: cascading failures, tipping points, spirals.
In other words

Testing a ship in a swimming pool and declaring it ocean-ready.

Four fundamental flaws. Not bugs — design choices.

One agent. Perfect rationality. No financial system. Linear math for a non-linear world.

The world needed a model that treats the economy as what it really is: millions of unique people making imperfect decisions, influencing each other, and creating outcomes no single equation can predict.

After 2008, the world's leading central banks asked a different question:

"What if we stopped averaging people — and started simulating them?"

They built a new kind of model.
Agent-Based Modeling
Section II
The Pioneers
Three central banks didn't wait for the next crisis.
They built the alternative.
BoE
Bank of England
2016 — First major central bank to adopt ABM
What they did
Built ABMs for housing markets, payment systems, and financial stability
With whom
INET Oxford & J. Doyne Farmer — complexity economics pioneer
Key result

ABMs captured emergent systemic risk and non-linear contagion — phenomena that DSGE structurally cannot produce

BoC
Bank of Canada — CANVAS
2022 — First inflation-targeting central bank to use ABM for forecasting
What they built
Heterogeneous households & firms with adaptive learning, not rational expectations
COVID-19 test
Successfully explained both the crash and recovery — DSGE could not
Key result

CANVAS outperforms DSGE in forecasting GDP growth and consumption — published in Journal of Economic Dynamics & Control, 2025

BoI
Bank of Italy — BeforeIT.jl
2025 — Open-source, anyone can use it
What they built
Full-country simulation: firms, households, banks, government — in Julia for speed
Why it matters
Open source — any central bank can adapt it to their own economy
Key result

First ABM to match the forecasting performance of traditional macroeconomic tools — European Economic Review, 2023

And they're not alone
0+ institutions now use agent-based models
ECB
European Central Bank
IMF
International Monetary Fund
BIS
Bank for Int'l Settlements
BBK
Deutsche Bundesbank
BdE
Bank of Spain
OeNB
Austrian National Bank
INET
Oxford Institute
SFI
Santa Fe Institute
The world's most respected financial institutions are building agent-based models.

The question is no longer "should we?" — it's "how fast can we start?"

But how does it actually work?
Section III
How It Works
Instead of one equation for the whole economy — simulate every participant individually.
Traditional approach
Y = f(x)
  • Write one equation for the whole economy
  • Solve it mathematically
  • Get one answer
  • Hope it matches reality
VS
Agent-based approach
  • Create thousands of agents with real behavior
  • Let them interact with each other
  • The economy emerges from their decisions
  • Run it 100 times — see the range of outcomes
How it works
Households
Banks
Traders
Migrants
Farmers
Every agent is unique
Each agent has its own income, savings, trust level, and behavior. A bazaar trader in Tashkent responds differently than a farmer in Fergana or a bank manager in the capital.
What makes it powerful
SHOCK
Decisions ripple through the system
When oil prices drop, migrant remittances fall. Households cut spending. Bazaar traders raise prices. Banks tighten credit. Each reaction triggers the next — exactly like a real economy.
Why central banks need this
Policy change
Rate +2%
→
Banks
Tighten lending
→
Households
Reduce spending
→
Bazaars
Prices adjust
→
Result
CPI, GDP, FX
Test policy before you implement it
Change the interest rate — and watch how every agent group responds. See which households are hurt. Which banks comply. How fast bazaar prices adjust. All before a single real decision is made.
This is no longer theoretical.

We built this for Uzbekistan.

217 agents. 11 groups. One economy.
Section IV
Built for Uzbekistan
Not a prototype. Not a concept.
A working behavioral simulator calibrated with real Uzbek data.
0 agents across 11 groups
50
Households
30
Bazaar Traders
25
Migrant Workers
20
Farmers
20
Businesses
15
Banks
15
Importers
12
Developers
12
Logistics
10
Foreign Investors
8
Government
Calibrated with real data
78%
of households save cash at home, not in banks
IFC / World Bank, 2020
82%
of bank assets are state-owned — rates don't follow policy signals
IMF FSAP, 2025
$14.8B
in remittances, 78% from Russia — a single-corridor vulnerability
IOM, 2025
What you can test
OIL
Oil Price Crash
Russia slows down, remittances collapse. Watch how households, traders, and banks react in sequence.
%
Rate Change
Raise or cut the policy rate. Watch which banks comply and which households feel it.
$
Dollarization Spiral
Currency panic triggers gold buying, dollar hoarding, and bazaar price spikes.
See it yourself
The simulator is live. Choose a scenario, watch 217 agents make decisions, and see how policy ripples through Uzbekistan's economy — in real time.
Launch the Simulator
The world's best central banks
are already simulating their economies.
Uzbekistan can lead Central Asia
in building the next generation of economic policy tools.
CBU Behavioral Macroeconomic Simulator