This is incremental time progression, not DES https://en.wikipedia.org/wiki/Discrete-event_simulation

We talk a lot about inference, but it’s less often we hear about its inverse, simulation

Simulation is not prediction

Generative models allow simulation

What model could have produced this data? [Infer]

↕

What data would be produced by this model? [Simulate]

def gen_initial():
  return {} # TODO: Set initial conditions
  
def get_sim_length():
  pass # Return a positive integer
  
def sim_next(history, t):
  pass # Look at all of history and simulate the next time step, which is t

def run_simulation():
  S = [gen_initial()]
  T = get_sim_length()
  for t in range(1, T+1):
    S.append(sim_next(S, t]))
  return pd.DataFrame(S) # Index is time

For example above: Walk w/ changing drift est. from data

For DES see this: https://www.youtube.com/watch?v=0KvA92ykPKI&ab_channel=UQComputingSociety

Possible example 1:

Idea: Get house data pres data with wiki package

Hypothesis: Midterms are bad for the current president; incumbents tend to win; non-incumbent races are a toss-up Use ABC! Find rules that agree with the actual path 80% of the time

Possible example 2: A VaR simulation problem, maybe on one of these data sets: https://github.com/lmc2179/lmc2179.github.io/blob/master/drafts/random_walk.md

Possible example 3: A problem requiring strong assumptions, like a model of how a firm grows; lets us see how sensitive we are the the assumptions

Possible example 4: A counterfactual example; how does the time series change if I intervene at time X

Bayesian bonus: Bayesian inference creates a natural way to do simulation too (link gelman)