Thanks for the explanations.
Actually 100 ppm/K is good enough for me, but I just want to know about the best, out of curiosity.
I need the variable resistor to work as a reference which is connected in serial with a sample. If an AC current of frequency w is passed through the sample and the variable resistor, a 3w signal will be generated on the sample due to heating effect. The variable resistor is adjusted to cancel the 1w signal so that the 3w signal on the sample can be detected. If you are familiar with this you will know that it is the so called 3-omega method. In this way, the reference variable resistor must have much smaller temperature coefficient than that of the sample.