Stochastic Approximation and Simulation using Running Median

Question

I have a function $$F(X_{t+1},Y_{t}^{med})= \alpha X_{t+1} + (1-\alpha) Y_{t}^{med},$$ where $$Y_{t}^{med} = Median(Y_1, Y_2,... Y_t).$$ Moreover, $Y_1 = X_1$ and $Y_t = F(X_{t},Y_{t-1}^{med})$ $\forall t>1$.

Suppose $X_t \sim X$ is a IID random variable following a Beta distribution, $X \sim Beta(2,1)$.

I wish to plot the evolution of $Y_{t}^{med}$ over time.

Currently, I have:

1) The function:

f[a_, {x_, y_}] := a x + (1 - a) y

2) Generate data. Here, I don't know how to generate the running median, that is $Y_{t}^{med}$ in Mathematica.

sim[length_] := 
 Module[{rv = RandomVariate[BetaDistribution[2, 1], length], y, yMed},
  y[1] = First@rv;
  yMed[t_Integer] := yMed[t] = MovingMedian[rv, t];
  y[t_Integer] := y[t] = f[0.5, {rv[[t]], yMed[t - 1]}];
  yMed /@ Range[length]]

3) Create the plot/graph for 100 iterations, 5 random walks:

ListPlot[Table[sim[100], {5}], PlotRange -> All, Joined -> True, 
 Frame -> False, AxesOrigin -> {0, 0}, AxesLabel -> {t, Y}, 
 GridLines -> {{}, {2/3}}, GridLinesStyle -> Directive[Gray, Dashed]]

Answer 1

Using

sim[length_] := 
 Module[{rv = RandomVariate[BetaDistribution[2, 1], length], y, yMed},
  y[1] = First@rv;
  yMed[t_Integer] := yMed[t] = Median[y /@ Range[t]];
  y[t_Integer] := y[t] = f[0.5, {rv[[t]], yMed[t - 1]}];
  yMed /@ Range[length]
 ]

results in plots like

Here Median[y /@ Range[t]] calculates the median for y[1] to y[t], as stated in your question.

MovingMedian is used to calculate the median values for a moving window with fixed length, e.g. the median for y[t-10] to y[t].
However, for such a simulation

sim2[length_Integer, medianW_Integer] := 
 Module[{rv = RandomVariate[BetaDistribution[2, 1], length], y, yMed},
 y[1] = First@rv;
 yMed[t_Integer] := 
 yMed[t] = Median[y /@ Range[Max[1, t - medianW], t]];
 y[t_Integer] := y[t] = f[0.5, {rv[[t]], yMed[t - 1]}];
 yMed /@ Range[length]]

ListPlot[Table[sim2[100, 10], {5}], PlotRange -> All, Joined -> True, 
 Frame -> False, AxesOrigin -> {0, 0}, AxesLabel -> {t, Y}, 
 GridLines -> {{}, {2/3}}, GridLinesStyle -> Directive[Gray, Dashed]]

would be the better implementation in your case.