David Chaston suggests that if you want to know where mortgage rates are headed next, watch what the big banks do with term deposit rates rather than where swap rates go

The number of borrowers 'underwater' would only ever be tiny if house prices fell -20% or more. Firstly, most mortgages require 20% equity and house prices haven't fallen anything like that. Secondly, banks all apply a 7.5% interest rate test to a new borrower's ability to pay higher rates from their income, even though they don't charge that. They make sure borrower income streams are resilient.

It seems that only an absolute maximum 4% of all mortgages could ever be under threat of being 'underwater'. These are those that were issued over 80% LVR in the past two years. And banks are likely to be very little exposed - they either require LMI (Lenders Mortgage Insurance) on those, or they self-insure LMI (and so specically provide for this risk). In the event, it would be less than 1% that might slip 'underwater'.

Why would banks arbitarily lower interest rates for every customer to protect a risk on a tiny fraction of them? Doesn't make sense.

The bank asset is the loan; the property is the security for that asset - splitting hairs, I know, but the last thing that banks want is their assets to be turned into 'property'.
That, to some degree, answers David' question above, "Why would they arbitrarity...?" Because they don't want to have to convert loans into property; neither do they want forced sales increasing, that might further damage the value of their security ( much lower property prices from negative feedback spiral). Much better to lower the price of loans across the board than risk contagion.
Lenders are continually Stress Testing their loan books, and they will know better than anyone what the trade-off between a possible increase in non-performing loans and lower interest rates is.

Really? Evidence? Borrowers break fixed rate mortgages all the time. We have no evidence this is happening any more now than usual.

We do have a proxy for this - the times our break fee calculator is used. Usage since December 2018 has been entirely normal. There was a spike in November, but it has died down now. Our calculator was launched in October 2017, so this data only goes back to then.

Here's the problem with that postulation. Ignoring (for the moment) mortgages for investors (the business of being a landlord), there are a bit more than $182.4 bln owed on mortgages on housing (RBNZ C22).

Those same owner-occupied houses are worth north of $812.9 bln. OK, lets say about half of them have no mortgage at all. So about $400 bln worth of houses are encumbered by mortgages worth $182 bln. That's an LVR of 45%. Lets be even more conservative and say a third of those have an LVR of under 25%. That leaves $270 bln carrying mortgages of at most $150 bln, or an LVR of 55%. 

Overall, even a -25% house price drop wouldn't affect many of these remaining ones in a way that would put them underwater. Sure, it would put more than currently under, but the proportion won't be large.

There are 1.75 mln private dwellings in New Zealand of which 1.1 mln are owner-occupied. If half have no mortgage, and only two thirds of the rest have an average of a 55% LVR, that means about 400,000 are exposed. Probably no more than 50,000 owner-occupied houses might be stressed (mostly Auckland, Wellington), and those that could face severe stress could perhaps number 5,000? Maybe 10,000. So that is a rounding error of the total houses - you would be lucky to get to 5% is a 'crisis' when house prices fell -25%, as you postulate.

And who really cares about landlord stress? If they had to sell up, that will just make those houses easier for other investors to buy or the tenants to afford to own. Their marklet will clear. The houses involved will still be there, probably occupied by the same people.