The routine may be called by the names g05xcf or nagf_rand_bb_inc_init.
3.1Brownian Bridge Algorithm
Details on the Brownian bridge algorithm and the Brownian bridge process (sometimes also called a non-free Wiener process) can be found in Section 2.6 in the G05 Chapter Introduction. We briefly recall some notation and definitions.
Fix two times and let be any set of time points satisfying . Let denote a -dimensional Wiener sample path at these time points, and let be any matrix such that is the desired covariance structure for the Wiener process. Each point of the sample path is constructed according to the Brownian bridge interpolation algorithm (see Glasserman (2004) or Section 2.6 in the G05 Chapter Introduction). We always start at some fixed point . If we set where is any -dimensional standard Normal random variable, then will behave like a normal (free) Wiener process. However if we fix the terminal value , then will behave like a non-free Wiener process.
The Brownian bridge increments generator uses the Brownian bridge algorithm to construct sample paths for the (free or non-free) Wiener process , and then uses this to compute the scaled Wiener increments
Such increments can be useful in computing numerical solutions to stochastic differential equations driven by (free or non-free) Wiener processes.
Conceptually, the output of the Wiener increments generator is the same as if g05xaf and g05xbf were called first, and the scaled increments then constructed from their output. The implementation adopts a much more efficient approach whereby the scaled increments are computed directly without first constructing the Wiener sample path.
Given the start and end points of the process, the order in which successive interpolation times are chosen is called the bridge construction order. The construction order is given by the array times. Further information on construction orders is given in Section 2.6.2 in the G05 Chapter Introduction. For clarity we consider here the common scenario where the Brownian bridge algorithm is used with quasi-random points. If pseudorandom numbers are used instead, these details can be ignored.
Suppose we require the increments of Wiener sample paths each of dimension . The main input to the Brownian bridge increments generator is then an array of quasi-random points where each point has dimension or depending on whether a free or non-free Wiener process is required. When g05xdf is called, the th sample path for is constructed as follows: if a non-free Wiener process is required set equal to the terminal value , otherwise construct as
where is the matrix described in Section 3.1. The array times holds the remaining time points in the order in which the bridge is to be constructed. For each set , find
and construct the point as
where or depending on whether a free or non-free Wiener process is required. The routine g05xef can be used to initialize the times array for several predefined bridge construction orders. Lastly, the scaled Wiener increments
Glasserman P (2004) Monte Carlo Methods in Financial Engineering Springer
1: – Real (Kind=nag_wp)Input
On entry: the starting value of the time interval.
2: – Real (Kind=nag_wp)Input
On entry: the end value of the time interval.
3: – Real (Kind=nag_wp) arrayInput
On entry: the points in the time interval at which the Wiener process is to be constructed. The order in which points are listed in times determines the bridge construction order. The routine g05xef can be used to create predefined bridge construction orders from a set of input times.
On exit: communication array, used to store information between calls to g05xdf. This array must not be directly modified.
6: – IntegerInput/Output
On entry: ifail must be set to , or to set behaviour on detection of an error; these values have no effect when no error is detected.
A value of causes the printing of an error message and program execution will be halted; otherwise program execution continues. A value of means that an error message is printed while a value of means that it is not.
If halting is not appropriate, the value or is recommended. If message printing is undesirable, then the value is recommended. Otherwise, the value is recommended. When the value or is used it is essential to test the value of ifail on exit.
On exit: unless the routine detects an error or a warning has been flagged (see Section 6).
6Error Indicators and Warnings
If on entry or , explanatory error messages are output on the current error message unit (as defined by x04aaf).
Errors or warnings detected by the routine:
On entry, and . Constraint: .
On entry, . Constraint: .
On entry, , and . Constraint: for all .
On entry, , for and . Constraint: all elements of times must be unique.
An unexpected error has been triggered by this routine. Please
See Section 7 in the Introduction to the NAG Library FL Interface for further information.
Your licence key may have expired or may not have been installed correctly.
See Section 8 in the Introduction to the NAG Library FL Interface for further information.
Dynamic memory allocation failed.
See Section 9 in the Introduction to the NAG Library FL Interface for further information.
8Parallelism and Performance
g05xcf is not threaded in any implementation.
The efficient implementation of a Brownian bridge algorithm requires the use of a workspace array called the working stack. Since previously computed points will be used to interpolate new points, they should be kept close to the hardware processing units so that the data can be accessed quickly. Ideally the whole stack should be held in hardware cache. Different bridge construction orders may require different amounts of working stack. Indeed, a naive bridge algorithm may require a stack of size or even , which could be very inefficient when is large. g05xcf performs a detailed analysis of the bridge construction order specified by times. Heuristics are used to find an execution strategy which requires a small working stack, while still constructing the bridge in the order required.
The following example program calls g05xafandg05xbf to generate two sample paths from a two-dimensional free Wiener process. It then calls g05xcfandg05xdf with the same input arguments to obtain the scaled increments of the Wiener sample paths. Lastly, the program prints the Wiener sample paths from g05xbf, the scaled increments from g05xdf, and the cumulative sum of the unscaled increments side by side. Note that the cumulative sum of the unscaled increments is identical to the output of g05xbf.