Such a rule of thumb would aways be specific to your game.
Looking at the numbers and setting the cost linearly can be very misleading. Some stats might be more useful than others. Some stats might have synergetic effects between each other. The usefulness of stats might not always be linear.
And then there might be game mechanics in your game which affect whether few strong units are better or worse than many weak units.
So the only way to really figure this out is through playtesting. Nothing beats playtesting with real human players. But in order to save time and money, you can also cut down the time you need by automatizing your playtesting.
Create a framework which allows you to quickly run a large number of automatic battle simulations with different pairings and tell you which side wins how often.
The most primitive simulation is to simulate simple unit vs. unit combat. Just simulate two units taking turns attacking each other until one is dead. Run that a thousand times, and you can see which unit wins how often. This should tell you which of the two units is superior and thus should cost more.
One further step is to simulate group combat. Extend your previous simulation framework to allow 1 vs multiple or even multiple vs. multiple units. This allows you to answer questions like "are 100 points worth of vampire bats stronger or weaker than 100 points worth of blood bats". But now you need to think about how units should pick their targets. Simulating various strategies and looking at their success rates can also provide you with useful data about which tactics players should choose in different situations. This data can also be very useful for your AI development.
The most comprehensive form of simulation is to run complete matches AI vs. AI. This allows you to run lots and lots of simulation series with slightly tweaked unit attributes and compare win/lose ratios. But this of course requires that you already have an AI for your game which is able to play in a way which is comparable to humans.
Still, in the end, no simulation can replace real playtesting. You players might figure out strategies you never thought about when developing your AI. Such a strategy might completely revolutionize the meta-game and force you to rebalance various units to make the game interesting again. No complex strategy game has really solved this problem yet. Even the most popular AAA strategy games keep releasing balance patches again and again to address balance problems discovered by the player community.
