It’s January 2018, which means that defaults must now be accounted for well in advance of their actual defaulting. In other words, loss of a financial instrument must now be recognised at its birth, death and every day in between. If that makes sense to you then just spare a moment’s thought about how it was done before today. In the old world (IAS39), the default of a financial instrument did not need to be considered until it actually defaulted. That also makes sense, except for one thing: provisions. You see, a provision is raised against all financial instruments just in case one or two of them go bad and then it’s OK, there’s no surprise, because we’ve already put aside some money for that. So in the old world, defaults would happen and if we provisioned correctly their impact wouldn’t be so bad because any loss would be netted away by the provision raised the month before. A result of this is that one could estimate next month’s losses by the amount of provision raised. Invert the recovery rate, and one could estimate the implied probability of default.
But what happens when your losses exceed your provisions? What if they exceed the provision by A LOT? From an accounting point of view that’s real bad. That’s what happened in the GFC. Defaults and losses spiked and the provisions that had been raised in the previous months leading up to the GFC were no where near large enough to absorb it.
One cannot just go and raise huge provisions either, that’s cheating. It’s also really bad for finance. Especially if it goes unutilised. The key is to estimate future losses by assuming that each and every instrument could default and go to loss at some point in its life, and couple this estimation with the health of the macro-economy.
Consider an economy with record low unemployment, record low inflation rates, and record high GDP. Assume also that these variables are strongly mean-reverting, which in real life they often are. Would you increase or decrease your provisions?
This is how we are to view the problem of loss provisioning and expected credit losses under IFRS 9, the new accounting standard, which must be implemented globally by today, January 2018.
Under IFRS 9, financial instruments are monitored over a grade of performing metrics right from day one of its life. If all the metrics are good, the loan is said to be in Stage 1. But just because it is in Stage 1 doesn’t mean it is immune to provisions. Stage 1 will have a provision raised against it, and it will equal the likelihood that each loan defaults at some point in the next 12 months. That’s a short period of time, and particularly if they are all good performing loans, that’s not much time for them to default, hence the PD for this cohort is usually very, very low. But crucially, non-zero.
Once any one or a combination of metrics crosses a pre-determined threshold, the instrument is considered Stage 2. This cohort of instruments contains loans which have satisfied early-warning indicators, suggesting that it has a considerably higher chance of defaulting. These loans are much closer to defaulting so their PDs are much higher. Furthermore, we adjust the time horizon to the instrument’s life (instead of just 12 months) allowing much more time for it to default, further increasing the PD.
Finally, as in IAS39, there is Stage 3, which contains all the instruments which have been observed as actually defaulting. These instruments attain a PD of 100% and have no future life.
Further penalties apply. As touched on earlier, the propensity for instruments to default are increased in macro-economies which display higher chances of deteriorating. Confusingly, default likelihoods actually increase when the key macro-economic variables are all good; and they are decreased when the same variables are really bad. Why is this?
Macro-variables can be shown to be strongly mean reverting. This means that when they are at very high or very low levels, they are more likely to snap back to their long-term averages. Thus, an economy with record low unemployment, record low inflation rates, and a record high GDP probably doesn’t much chance of getting any better. As a result, those same variables are likely to be at worse positions in 12 months time and we account for this by scaling up the default and loss rates. Conversely, an economy with record high unemployment, record high inflation and low GDP will probably be in a better position in 12 months time, so the default and loss rates are reduced.
An economy with all key variables already at the mean suffers as well, because it is at this point where we have the least confidence of any sort of direction. Low statistical confidence also hurts the forecast by raising default and loss rates.