Outline of what needs to happen to get a working paper out of this. #4

I need to figure out and explain:

The economic goals
The coding goals
The literature goals
What I'm still unsure about/need to understand better.

What I am still unsure about.

Criticizm from my previous presentation

Why isn't this a representative agent model?
Why don't I use Rienforcemnt learning?
I used the wrong notation for a markov system (what about a non-markov system?)

Concerns I have

What types of nash equilibria represent the situation?
- I have many discrete agents
If there are multiple pure equilibria there might be mixed strategy equilibria
Does ML have anything on nash equbria that I'm not aware of?
How do I handle an arbitrary number of agents? Is there a way to include a "non-entry" option? Maybe negative numbers of satellites?

What am I going to do about it?

Other random thoughts

What if the state space is "Here are mine" and "here is the general other state", where we have a set of standardized orbits? That doesn't represent the real world that well though. And there are national services handling

I need to figure out and explain: - The economic goals - The coding goals - The literature goals - What I'm still unsure about/need to understand better. # What I am still unsure about. Criticizm from my previous presentation - Why isn't this a representative agent model? - Why don't I use Rienforcemnt learning? - I used the wrong notation for a markov system (what about a non-markov system?) Concerns I have - What types of nash equilibria represent the situation? - I have many discrete agents - If there are multiple pure equilibria there might be mixed strategy equilibria - Does ML have anything on nash equbria that I'm not aware of? - How do I handle an arbitrary number of agents? Is there a way to include a "non-entry" option? [Maybe negative numbers of satellites?](https://datascience.stackexchange.com/questions/96933/how-to-handle-undefined-or-null-data-in-a-neural-network) # What am I going to do about it? # Other random thoughts What if the state space is "Here are mine" and "here is the general other state", where we have a set of standardized orbits? That doesn't represent the real world that well though. And there are national services handling

Other concerns.

After a year+ of not looking at it, I am missing:

The julia code I had to run the system.
The raw presentation I gave.
Any clue of how to handle game theory aspect

What I have

Notes on what I was doing
Literature and Lit review
Experience
Improvements in pytorch world.

# Other concerns. After a year+ of not looking at it, I am missing: - The julia code I had to run the system. - The raw presentation I gave. - Any clue of how to handle game theory aspect What I have - Notes on what I was doing - Literature and Lit review - Experience - Improvements in pytorch world.

What I need:

Gameplan

What question I am going to answer
What tools I am going to use

# What I need: Gameplan - [ ] What question I am going to answer - [ ] What tools I am going to use

PLAN to get back on track

Review What I wrote
Review Lit
Investigate a simple question - How to extend models handle two different types of agents (Market vs Gov)?
Review previous code
Examine different approaches
Make a plan.
Execute
????
Profit

# PLAN to get back on track - [x] Review What I wrote - [ ] Review Lit - [ ] Investigate a simple question - How to extend models handle two different types of agents (Market vs Gov)? - [ ] Review previous code - [ ] Examine different approaches - [ ] Make a plan. - [ ] Execute - [ ] ???? - [ ] Profit

What if I examine a case with 2 satellite operators and two competing military satellite systems?

Are Satellite lifetimes, survival, or debris generation ergodic? If not, then we have an issue with standard approaches.

Got my julia code and writing back (it was sitting in the temporaryWork branch on Bitbucket).

I think I am going to continue with what I was doing, but limit it to simple situations.

Got my julia code and writing back (it was sitting in the `temporaryWork` branch on Bitbucket). I think I am going to continue with what I was doing, but limit it to simple situations.

So I have determined to limit this to a cournot model with 2 internet satellite providers in a bandwidth market. Why? Because it is a simple model of strategic interaction that allows for extensions into repeated investments.

I am not going to include government interactions until I know more about how to model those interactions.

So I have determined to limit this to a cournot model with 2 internet satellite providers in a bandwidth market. Why? Because it is a simple model of strategic interaction that allows for extensions into repeated investments. I am not going to include government interactions until I know more about how to model those interactions.

Consider

The physical dynamics are a markov process
Firms compete in each stage according to a cournot game (adding bandwidth capacity by adding satellites)

Only consider two competitors in a space.

Solve this computationally and manually if possible.
Notes are all at home.

This gives two bellman equations that are solved together.

Consider - The physical dynamics are a markov process - Firms compete in each stage according to a cournot game (adding bandwidth capacity by adding satellites) Only consider two competitors in a space. Solve this computationally and manually if possible. Notes are all at home. This gives two bellman equations that are solved together.

Simple pricing function to represent bandwidth


P(Q) = a - bQ

Simple pricing function to represent bandwidth \[ P(Q) = a - bQ \]

Based on Reinforcement Learning and Stochastic Optimization's definition, this is a multiagent, state dependent, Policy Function Approximation problem. If I care about value of entry problems, then there is a lookahead (Value function approximation) portion.

Labels Milestones