That's true if, and only if, the expected deflation either changes or the actual deflation doesn't well match the predicted deflation. I agree that you can get all kinds of strange and bad results if you try to denominate an agreement in an unstable or unpredictable currency.
It's true even if expected deflation doesn't change and actual deflation matches the predicted deflation well; in fact I was assuming this would probably be the case. The reduction in uncertainty about deflation and other factors over the time period in question is what causes the value of the coins to increase.
Suppose you borrow money for (say) two years. At the start of those two years people can only estimate deflation over the next few years rather than know for certain, and the further out they look the less accurately they can estimate. This uncertainty affects how much of the anticipated effects of deflation people are willing to take into account. In two years' time everyone will pretty much know what happened over those two years and will be in a better position to judge what will happen in the third and subsequent years, so more of the effect of deflation will be priced in.
Exactly. That's why you don't want to denominate an agreement in a currency that isn't very predictable over the time frame of the agreement. Otherwise, people will become very risk averse and tend to prefer shorter term agreements because it costs money to manage risk. It would be madness to offer or accept a 30 year mortgage denominated in bitcoins today.
It doesn't matter much how predictable the currency is - as long as there's the slightest sliver of uncertainty about the future (and who can really say for sure what will happen in a thousand years or a million?) you will have this reduction in uncertainty over time causing more deflation to be priced in. I don't think the actual level of uncertainty even affects the rate at which deflation will be priced in on average over the long term - only changes to it matter. For example, suppose we have the following incredibly simplified situation where we're at the end of some year
y, we know statement S0 is true and are uncertain about the rest of the statements S1, S2, ...
S0: The economy behaved the same in year
y as in all previous years, with the same growth rate and the same 5% deflation, and it's 99.99 percent certain that a year later we'll confirm statement S1 is true.
S
n: The economy behaved the same in year
y+n as in all previous years, with the same growth rate and the same 5% deflation, and it's 99.99 percent certain that a year later we'll confirm statement S
(n+1) is true.
Up until the point one of the statements S
n is discovered to be false - which is by our assumptions impossible to predict in advance - all the years are identical and in every year we have identical expectations for the future, so the effect of expected deflation must necessarily be the same in each year and cancel out, making it impossible for that to compensate for the effect of the deflation on loans at all. This works because as the loan period passes we're more certain about deflation and the economy not just in that year but all the way into the distant future. You can change that 99.99 percent to 99.99999999 percent and the argument still works.
Here's another example of the same error where it's easier to see: "Who would want to borrow money with an inflationary currency? The mortgage rate would likely have to be adjustable and could climb up really high, forcing you to pay back lots more money."
That's not even close to the same statement. There is actually a fairly direct deflationary equivalent of this problem - unexpectedly high rates of price deflation could force you to pay back lots more than you were expecting to - but I'm just talking about the predictable, routine increase in the amount you owe in real terms.