||You can't manufacture data you don't have, Jakub, but you can fake it.
There are several algorithms for "pseudo natural color" around. One that is described clearly appears at http://myweb.facstaff.wwu.edu/wallin/envr442/COB_ColorIR2004_Processing.pdf . (Please ignore the AML code there: it has typos.) Essentially, it estimates what the blue band might look like based on the other bands. You can implement formulae like these with raster calculations. Let the SPOT bands be [B1] (500 - 590 nm), [B2] (610 - 680 nm, same as the "panchromatic" band), and [B3] (780 - 890 nm). (Band B4, 1580 - 1750 nm, is too far from the visible spectrum to be much use normally.) Compute
Blue = 0.85 * [B1] - 0.15 * [B3]
Green = 0.65 * [B1] + 0.35 * [B3]
Red = [B2]
and put them into a three-band "stack."
I mistrust this particular algorithm (the blue will never be intense and can even become negative), but it does show you the flavor of this approach.
More sophisticated methods are referenced at http://ceos.cnes.fr:8100/cdrom-00b/ceos1/science/gdta/ang/a1an/82.htm . The better ones require having true color images on hand for calibrating the estimates using some kind of classification or regression method. I could not find anything on the SPOT site itself that gives any information about their "pseudo natural color" product, even though some of their technical documents refer to it.