Bootstrap sampling question (incorporating with higher moments?)

I currently have an issue. I developed an asset allocation model based on some historical data, results were above average and now I want to confirm those results by doing a bootstrap sampling.

However, the asset allocation model im working with is based on mean, variance, skewness, and kurtosis.

If i do the bootstrap method provided in matlab,

the

    [bootstat,bootsam] = bootstrp(10000 ....

I do gain 10000 timeseries of returns, but the 3rd and 4th moment are all over the place. For example my kurtosis varies between -1.5 and -0.5. Which is not even close to the kurtosis of the historical data.

Can someone help me?

[Tom Lane]

Could you clarify what you are doing? If I generate x either from rand or from trnd(3,...) I get very different kurtosis values from them. But if I use bootstrp(10000,@kurtosis,x), in each case I get a mean value that is very similar to the sample kurtosis of x.