/

An analytical/semi-analytical analysis of the characteristics of the Expected Worst Loss in T realisations for Normal random variables

[this page | pdf | back links]

Suppose we have  which are independent identically distributed Normal random variables, i.e. . The worst loss in  realisations is .

 

What, in such circumstances is the expected worst loss (in  realisations)?

 

Without much loss of generality, we can focus on , i.e. unit Normal random variables, as .

 

If  then all of the  must also satisfy . As the  are independent and identically distribution, this means that:

 

 

where  is the cumulative distribution function of the unit Normal distribution.

 

So the probability density function of  is  where:

 

 

The expected value of , i.e. the expected worst loss in  realisations for a unit Normal distribution, is then , i.e.:

 

 

Using a symbolic algebra engine we find that there are analytical formulae  for  = 1, 2 or 3 but not thereafter:

 

1

2

3

 

As  increases   becomes more negative, but only relatively slowly:

 

1

0

 

2

-0.5641895835

-0.5641895835

4

-1.029375373

-0.4651857895

8

-1.423600306

-0.394224933

16

-1.765991393

-0.342391087

32

-2.069668828

-0.303677435

64

-2.343733465

-0.274064637

128

-2.594597369

-0.250863904

256

-2.826863279

-0.232265910

 

We may check that the values shown above are reasonable using a simulation approach, e.g. using VBA code in Microsoft Excel as per VBA code that can be used to check this analysis.

 

The worst loss can also be thought of as a specific quantile of a sample of Normal random variables. It might be viewed as the ’th quantile (i.e. half way between 0 and , given that the sample has  observations) . We might therefore expect its expected value to be similar to the corresponding quantile point of the Normal distribution. However, the Normal probability density function is not normally approximately flat at the relevant quantile point (and is instead upward sloping), so the Expected Worst Loss in T realisations is normally somewhat above the ’th Normal quantile point:

 

c.f.

1

0

0

2

-0.5641895835

-0.67449

4

-1.029375373

-1.15035

8

-1.423600306

-1.53412

16

-1.765991393

-1.86273

32

-2.069668828

-2.15387

64

-2.343733465

-2.41756

128

-2.594597369

-2.66007

256

-2.826863279

-2.88563

 


Desktop view | Switch to Mobile