Hi,I am working on earnings/cashflow at risk numbers over a long period for a USD interest rate risk portfolio. The problem that I have is that the distribution of paths of the short term USD interest rate is unsatisfactory using a constant log normal volatility for the rate. A high number leads to excess volatility when rates increase (I am using the market forward curve as a drift). A low volatility leads to insufficient volatility at the start of the simulation.What method should I use to avoid this problem. Should I make volatility a decreasing function over time or a function of the short term interest rate? Or should I use a normal volatility with the constraint that the rate is never less than zero.Thanks

Many people compromise by using a CEV model: dr = const * r^b * dW = const / r^(1-b) * r * dWExponent b=1/2 is always popular.Others use a normal model, but may (or may not) put a reflecting barrier in. Many people think the barrier should be significantly below r=0, possibly r = -50bps or so.

Or should I use a normal volatility with the constraint that the rate is never less than zero.Yes and No.Yes, the volatility is more normal than log normal. Try some statistical tests.No, the interest rate can go less than zero. Japan??