Stochastic Calculus

Lesson 4 of 11

The Itô Integral

Left-endpoint and right-endpoint sums for ∫W dW converge to different answers, ½W_T² − ½T and ½W_T² + ½T, and finance takes the left one because a position has to be set before the price moves.

Integrating against a path with no slope

The ordinary integral adds up f(t) Δtf(t)\,\Delta t over the pieces of an interval. An integral against a curve gg weighs each piece by how much gg moved instead:

∫0Tf dg=lim⁡∑if(ti) Δgi,Δgi=g(ti+1)−g(ti).\int_0^T f\,dg = \lim \sum_i f(t_i)\,\Delta g_i, \qquad \Delta g_i = g(t_{i+1}) - g(t_i).

If gg is smooth, Δgi≈g′(ti) Δt\Delta g_i \approx g'(t_i)\,\Delta t, so the integral against gg becomes an ordinary integral over time, ∫0Tf(t) g′(t) dt\int_0^T f(t)\,g'(t)\,dt. That step needs a slope.

Take g(t)=1.068sin⁡(πt/2)g(t) = 1.068 \sin(\pi t / 2) and zoom in around t=0.6t = 0.6 with windows of width 1,14,116,164,12561, \tfrac14, \tfrac1{16}, \tfrac1{64}, \tfrac1{256}. The rise over run inside the window (the range of the curve divided by the window width) settles on 0.990.99, which is g′(0.6)=1.068⋅π2cos⁡(0.3π)≈0.99g'(0.6) = 1.068 \cdot \tfrac{\pi}{2}\cos(0.3\pi) \approx 0.99. The curve straightens into a line.

Do the same to a Brownian path ending at the same value, W(1)=1.068W(1) = 1.068. The rise over run at the five zoom levels is 1.7,3.1,3.5,6.9,13.91.7, 3.1, 3.5, 6.9, 13.9. Over a window of width hh the path moves about h\sqrt h, so the ratio grows like 1/h1/\sqrt h and roughly doubles with every fourfold zoom. Every window looks as jagged as the last, and there is no W′(t)W'(t) to substitute. So ∫W dW\int W\,dW cannot be rewritten as ∫W W′ dt\int W\,W'\,dt; it has to be defined directly as a limit of sums ∑W ΔW\sum W\,\Delta W, and the next question is where in each piece to read WW.