Editing Monte Carlo methods in finance (section)

=== Variance reduction ===
Square root convergence is slow, and so using the naive approach described above requires using a very large number of sample paths (1 million, say, for a typical problem) in order to obtain an accurate result. Remember that an estimator for the price of a derivative is a random variable, and in the framework of a risk-management activity, uncertainty on the price of a portfolio of derivatives and/or on its risks can lead to suboptimal risk-management decisions.

This state of affairs can be mitigated by [[variance reduction]] techniques.

==== [[Antithetic variates|Antithetic paths]] ====
A simple technique is, for every sample path obtained, to take its antithetic path &mdash; that is given a path <math>\{\varepsilon_1,\dots,\varepsilon_M\}</math> to also take <math>\{-\varepsilon_1,\dots,-\varepsilon_M\}</math>. Since the variables <math>\varepsilon_i</math> and <math>-\varepsilon_i</math> form an antithetic pair, a large value of one is accompanied by a small value of the other. This suggests that an unusually large or small output computed from the first path may be balanced by the value computed from the antithetic path, resulting in a reduction in variance.<ref>{{Cite book|title = Monte Carlo methods in financial engineering|url = https://archive.org/details/montecarlomethod53glas|url-access = limited|last = Glasserman|first = P.|publisher = Springer.|year = 2004|location = New York|pages = [https://archive.org/details/montecarlomethod53glas/page/n213 205]| isbn=9780387004518 }}</ref> Not only does this reduce the number of normal samples to be taken to generate ''N'' paths, but also, under same conditions, such as negative correlation between two estimates, reduces the variance of the sample paths, improving the accuracy.

==== [[Control variates|Control variate method]] ====
It is also natural to use a [[control variate]]. Let us suppose that we wish to obtain the Monte Carlo value of a derivative ''H'', but know the value analytically of a similar derivative I. Then ''H''* = (Value of ''H'' according to Monte Carlo) + B*[(Value of ''I'' analytically) &minus; (Value of ''I'' according to same Monte Carlo paths)] is a better estimate, where B is covar(H,I)/var(H).

The intuition behind that technique, when applied to derivatives, is the following: note that the source of the variance of a derivative will be directly dependent on the risks (e.g. delta, vega) of this derivative. This is because any error on, say, the estimator for the forward value of an underlier, will generate a corresponding error depending on the delta of the derivative with respect to this forward value. The simplest example to demonstrate this consists in comparing the error when pricing an at-the-money call and an at-the-money straddle (i.e. call+put), which has a much lower delta.

Therefore, a standard way of choosing the derivative ''I'' consists in choosing a [[replicating portfolio]]s of options for ''H''. In practice, one will price ''H'' without variance reduction, calculate deltas and vegas, and then use a combination of calls and puts that have the same deltas and vegas as control variate.

==== Importance sampling ====
[[Importance sampling#Application to simulation|Importance sampling]] consists of simulating the Monte Carlo paths using a different probability distribution (also known as a change of measure) that will give more likelihood for the simulated underlier to be located in the area where the derivative's payoff has the most convexity (for example, close to the strike in the case of a simple option). The simulated payoffs are then not simply averaged as in the case of a simple Monte Carlo, but are first multiplied by the likelihood ratio between the modified probability distribution and the original one (which is obtained by analytical formulas specific for the probability distribution). This will ensure that paths whose probability have been arbitrarily enhanced by the change of probability distribution are weighted with a low weight (this is how the variance gets reduced).

This technique can be particularly useful when calculating risks on a derivative. When calculating the delta using a Monte Carlo method, the most straightforward way is the ''black-box'' technique consisting in doing a Monte Carlo on the original market data and another one on the changed market data, and calculate the risk by doing the difference. Instead, the importance sampling method consists in doing a Monte Carlo in an arbitrary reference market data (ideally one in which the variance is as low as possible), and calculate the prices using the weight-changing technique described above. This results in a risk that will be much more stable than the one obtained through the ''black-box'' approach.