Ideally, I would like to kill two birds with one stone.
1st bird: leaving keys completely exposed to sign blocks is too risky.
2nd bird: (unrelated to proof of stake) Keeping your money in a hot wallet is too risky. Keeping your money in a cold wallet is too inconvenient. Constantly rebalancing between your hot wallet and your cold wallet is also too inconvenient. Ideally, one wallet should be able to perform both hot and cold functions.
My idea for the stake key was bad because it wasn't useful as a hot wallet. 0.01% is too small to be usefully spent in most cases. Here is a new idea.
Suppose that 5% loss is an acceptable risk level for a one-off theft of the stake key.
How about the following txn rule for the stake key:
Amount Returned in Change to Public Key >= all coins sent to other addresses * { max (k, k*(t/accumulated coin-age on public key) }
k and t are positive constants. Coin age is measured in years. So if k=19 and t=1/12 (i.e. one month), ...
You can spend up to 5% of the a mature balance immediately and then about 0.001% per block afterwards. Theft loss is limited at close to 5%.
I think 5% will often allow you to use the key as a mixed hot/cold wallet.
Another nice thing is that the risk is self-limiting. 5% is fine for small balances but likely too much for large balances.
If you have a huge balance and mine PoS blocks less than once a month your risk will be less than 5%. Thus keeping a large balance online all the time is less risky. Bringing it online occasionally is more risky. This is very good for network security.
(e.g. if you mine PoS blocks once per day then your theft risk will be limited to about 0.166%; if you have the same balance and mine once per month, your theft risk increases to 5%.)
It would be nice if the parameters k and t could be user-specified (e.g. listed in the public key address (e.g. PCpzxSXGBLVuLEr2EAyxEnQH1QYvUKN9kuinsert k and t here)
They would have to have some minimum values to prevent stake signers from protecting themselves from too much risk. (i.e. if k>19 or t>1/12 then the key can't be used to sign for PoS)
Such a key could still be useful as a hot wallet though.
If you were running a business then you would want a very low value for t. That would allow your spending capabilities to recharge quickly, but keep your ability to spend per day limited.
[Edit: Fixed some algebraic errors]
I fail to see what benefits this offers. Someone deposits 5k BTC to buy MPOE bonds. 100% of the 5k should be locked somewhere until the next (last Friday of a month). At that point up to 100% of the 5k should be available, in case the person wants to cash in. Your proposed system serves neither of these, paper wallets do.