Tag Archive

You are currently browsing the tag archive for the ‘ergodic’ tag.

Learning the ergodic decomposition

September 27, 2014 in Decision theory, Probability, statistics | Tags: ergodic, learning, merging | by Eran | 1 comment

This post describes the main theorem in my new paper with Nabil. Scroll down for open questions following this theorem. The theorem asserts that a Bayesian agent in a stationary environment will learn to make predictions as if he knew the data generating process, so that the as time goes by structural uncertainty dissipates. The standard example is when the sequence of outcomes is i.i.d. with an unknown parameter. As times goes by the agent learns the parameter.

The formulation of `learning to make predictions’ goes through merging, which traces back to Blackwell and Dubins. I will not give Blackwell and Dubin’s definition in this post but a weaker definition, suggested by Kalai and Lehrer.

A Bayesian agent observes an infinite sequence of outcomes from a finite set ${A}$ . Let ${\mu\in\Delta(A^\mathbb{N})}$ represent the agent’s belief about the future outcomes. Suppose that before observing every day’s outcome the agent makes a probabilistic prediction about it. I denote by ${\mu(\cdot|a_0,\dots,a_{n-1})}$ the element in ${\Delta(A)}$ which represents the agent’s prediction about the outcome of day ${n}$ just after he observed the outcomes ${a_0,\dots,a_{n-1}}$ of previous days. In the following definition it is instructive to think about ${\tilde\mu}$ as the true data generating process, i.e., the process that generates the sequence of outcomes, which may be different from the agent’s belief.

Definition 1 (Kalai and Lehrer) Let ${\mu,\tilde\mu\in\Delta(A^\mathbb{N})}$ . Then ${\mu}$ merges with ${\tilde\mu}$ if for ${\tilde\mu}$ -almost every realization ${(a_0,\dots,a_{n-1},\dots)}$ it holds that

$\displaystyle \lim_{n\rightarrow\infty}\|\mu(\cdot|a_0,\dots,a_{n-1})-\tilde\mu(\cdot|a_0,\dots,a_{n-1})\|=0.$

Assume now that the agent’s belief ${\mu}$ is stationary, and let ${\mu=\int \theta~\lambda(\mathrm{d}\theta)}$ be its ergodic decomposition. Recall that in this decomposition ${\theta}$ ranges over ergodic beliefs and ${\lambda}$ represents structural uncertainty. Does the agent learn to make predictions ? Using the definition of merging we can ask, does ${\mu}$ merges with ${\theta}$ ? The answer, perhaps surprisingly, is no. I gave an example in my previous post.

Let me now move to a weaker definition of merging, that was first suggested by Lehrer and Smorodinsky. This definition requires the agent to make correct predictions in almost every period.

Definition 2 Let ${\mu,\tilde\mu\in\Delta(A^\mathbb{N})}$ . Then ${\mu}$ weakly merges with ${\tilde\mu}$ if ${\tilde\mu}$ -almost every realization ${(a_0,\dots,a_{n-1},\dots)}$ it holds that

$\displaystyle \lim_{n\rightarrow\infty,n\in T}\|\mu(\cdot|a_0,\dots,a_{n-1})-\tilde\mu(\cdot|a_0,\dots,a_{n-1})\|=0$

for a set ${T\subseteq \mathbb{N}}$ of periods of density ${1}$ .

The definition of weak merging is natural: patient agents whose belief weakly merges with the true data generating process will make almost optimal decisions. Kalai, Lehrer and Smorodinsky discuss these notions of mergings and also their relationship with Dawid’s idea of calibration.

I am now in a position to state the theorem I have been talking about for two months:

Theorem 3 Let ${\mu\in\Delta(A^\mathbb{N})}$ be stationary, and let ${\mu=\int \theta~\lambda(\mathrm{d}\theta)}$ be its ergodic decomposition. Then ${\mu}$ weakly merges with ${\theta}$ for ${\lambda}$ -almost every ${\theta}$ .

In words: An agent who has some structural uncertainty about the data generating process will learn to make predictions in most periods as if he knew the data generating process.

Finally, here are the promised open questions. They deal with the two qualification in the theorem. The first question is about the “ ${\lambda}$ -almost every ${\theta}$ ” in the theorem. As Larry Wasserman mentioned this is unsatisfactory in some senses. So,

Question 1 Does there exists a stationary ${\mu}$ (equivalently a belief ${\lambda}$ over ergodic beliefs) such that ${\mu}$ weakly merges with ${\theta}$ for every ergodic distribution ${\theta}$ ?

The second question is about strengthening weak merging to merging. We already know that this cannot be done for arbitrary belief ${\lambda}$ over ergodic processes, but what if ${\lambda}$ is concentrated on some natural family of processes, for example hidden markov processes with a bounded number of hidden states ? Here is the simplest setup for which I don’t know the answer.

Question 2 The outcome of the stock market at every day is either U or D (up or down). An agent believes that this outcome is a stochastic function of an unobserved (hidden) state of the economy which can be either G or B (good or bad): When the hidden state is B the outcome is U with probability ${q_B}$ (and D with probability ${1-q_B}$ ), and when the state is G the outcome is U with probability ${q_G}$ . The hidden state changes according to a markov process with transition probability ${\rho(B|B)=1-\rho(G|B)=p_B}$ , ${\rho(B|G)=1-\rho(G|G)=p_G}$ . The parameter is ${(p_B,p_G,q_B,q_G)}$ and the agent has some prior ${\lambda}$ over the parameter. Does the agent’s belief about outcomes merge with the truth for ${\lambda}$ -almost every ${(p_B,p_G,q_B,q_G)}$ ?.

Structural uncertainty

July 29, 2014 in Decision theory, Probability | Tags: bayesianism, ergodic, uncertainty | by Eran | Comments closed

Four agents are observing infinite streams of outcomes in ${\{S,F\}}$ . None of them knows the future outcomes and as good Bayesianists they represent their beliefs about unknowns as probability distributions:

Agent 1 believes that outcomes are i.i.d. with probability ${1/2}$ of success.
Agent 2 believes that outcomes are i.i.d. with probability ${\theta}$ of success. She does not know ${\theta}$ ; She believes that ${\theta}$ is either ${2/3}$ or ${1/3}$ , and attaches probability ${1/2}$ to each possibility.
Agent 3 believes that outcomes follow a markov process: every day’s outcome equals yesterday’s outcome with probability ${3/4}$ .
Agent 4 believes that outcomes follow a markov process: every day’s outcome equals yesterday’s outcome with probability ${\theta}$ . She does not know ${\theta}$ ; Her belief about ${\theta}$ is the uniform distribution over ${[0,1]}$ .

I denote by ${\mu_1,\dots,\mu_4\in\Delta\left(\{S,F\}^\mathbb{N}\right)}$ the agents’ beliefs about future outcomes.

We have an intuition that Agents 2 and 4 are in a different situations from Agents 1 and 3, in the sense that are uncertain about some fundamental properties of the stochastic process they are facing. I will say that they have `structural uncertainty’. The purpose of this post is to formalize this intuition. More explicitly, I am looking for a property of a belief ${\mu}$ over ${\Omega}$ that will distinguish between beliefs that reflect some structural uncertainty and beliefs that don’t. This property is ergodicity.

Definition 1 Let ${\zeta_0,\zeta_1,\dots}$ be a stationary process with values in some finite set ${A}$ of outcomes. The process is ergodic if for every block ${\bar a=(a_0,\dots,a_{k})}$ of outcomes it holds that

$\displaystyle \begin{array}{rcl} \lim_{n\rightarrow\infty}&\frac{\#\left\{0\le t < n|\zeta_t=a_0,\dots,\zeta_{t+k}=a_{k}\right\}}{n}=\\&\mathop{\mathbb P}(\zeta_0=a_0,\dots,\zeta_{k}=a_{k})~\text{a.s}.\end{array}$

A belief ${\mu\in \Delta(A^\mathbb{N})}$ is ergodic if it is the distribution of an ergodic process

Before I explain the definition let me write the ergodicity condition for the special case of the block ${\bar a=(a)}$ for some ${a\in A}$ (this is a block of size 1):

$\displaystyle \lim_{n\rightarrow\infty}\frac{\#\left\{0\le t < n|\zeta_t=a\right\}}{n}=\mathop{\mathbb P}(\zeta_0=a)~\text{a.s}.\ \ \ \ \ (1)$

In the right side of (1) we have the (subjective) probability that on day ${0}$ we will see the outcome ${a}$ . Because of stationarity this is also the belief that we will see the outcome ${a}$ on every other day. In the left side of (1) we have no probabilities at all. What is written there is the frequency of appearances of the outcome ${a}$ in the realized sequence. This frequency is objective and has nothing to do with our beliefs. Therefore, the probabilities that a Bayesian agent with ergodic belief attaches to observing some outcome is a number that can be measured from the process: just observe it long enough and check the frequency in which this outcome appears. In a way, for ergodic processes the frequentist and subjective interpretations of probability coincide, but there are legitimate caveats to this statement, which I am not gonna delve into because my subject matter is not the meaning of probability. For my purpose it’s enough that ergodicity captures the intuition we have about the four agents I started with: Agents 1 and 3 both give probability ${1/2}$ to success in each day. This means that if they are sold a lottery ticket that gives a prize if there is a success at day, say, 172, they both price this lottery ticket the same way. However, Agent 1 is certain that in the long run the frequency of success will be ${1/2}$ . Agent 2 is certain that it will be either ${2/3}$ or ${1/3}$ . In fancy words, ${\mu_1}$ is ergodic and ${\mu_2}$ is not.

So, ergodic processes capture our intuition of `processes without structural uncertainty’. What about situations with uncertainty ? What mathematical creature captures this uncertainty ? Agent 2’s uncertainty seems to be captured by some probability distribution over two ergodic processes — the process “i.i.d. ${2/3}$ ” and the process “i.i.d. ${1/3}$ ”. Agent 2 is uncertain which of these processes he is facing. Agent 4’s uncertainty is captured by some probability distribution over a continuum of markov (ergodic) processes. This is a general phenomena:

Theorem 2 (The ergodic decomposition theorem) Let ${\mathcal{E}}$ be the set of ergodic distributions over ${A^\mathbb{N}}$ . Then for every stationary belief ${\mu\in\Delta(A^\mathbb{N})}$ there exists a unique distribution ${\lambda}$ over ${\mathcal{E}}$ such that ${\mu=\int \theta~\lambda(\text{d}\theta)}$ .

The probability distribution ${\lambda}$ captures uncertainty about the structure of the process. In the case that ${\mu}$ is an ergodic processes ${\lambda}$ is degenerated and there is no structural uncertainty.

Two words of caution: First, my definition of ergodic processes is not the one you will see in textbooks. The equivalence to the textbook definition is an immediate consequence of the so called ergodic theorem, which is a generalization of the law of large numbers for ergodic processes. Second, my use of the word `uncertainty’ is not universally accepted. The term traces back at least to Frank Knight, who made the distinction between risk or “measurable uncertainty” and what is now called “Knightian uncertainty” which cannot be measured. Since Knight wrote in English and not in Mathematish I don’t know what he meant, but modern decision theorists, mesmerized by the Ellsberg Paradox, usually interpret risk as a Bayesian situation and Knightian uncertainty, or “ambiguity”, as a situation which falls outside the Bayesian paradigm. So if I understand correctly they will view the situations of these four agents mentioned above as situations of risk only without uncertainty. The way in which I use “structural uncertainty” was used in several theory papers. See this paper of Jonathan and Nabil. And this and the paper which I am advertising in these posts, about disappearance of uncertainty over time. (I am sure there are more.)

To be continued…

Tag Archive

Learning the ergodic decomposition

Structural uncertainty

Email Subscription

Entries by:

Recent Posts

Recent Comments

Blogroll

Archives

Meta