Page 98 - Annual Report 2023

29 BANKA KOMBËTARE TREGTARE
Notes to the Consolidated Financial Statements for the year ended 31 December 2023
(amounts in USD, unless otherwise stated)

Significant Deterioration through relative threshold
The bank computes a relative threshold matrix that gives for each rating a prediction of what the expected rating for each time horizon
should be.
Through-the-cycle (TTC) transition matrices give the percentage of counterparties which moved from one rating to another over a
specific time interval. TTC matrices over a 10 year horizon are taken since this gives a comfortable horizon for a relative threshold.

Computing relative thresholds for SICR
Given that the Bank does not have sufficient internal historical data to compile transition matrices, the bank relied on proxies provided
by external credit rating agencies. External TTC transition matrices for all European/World entities provided by international credit rating
agencies are used. The following steps are then carried out:
First the mapping between the internal and external rating systems is performed;
From there, a TTC transition matrix with the number of observations (number of entities that changed from one specific rating to another)
is used and a weighted average per mapped rating is computed in order to compute the internal ratings TTC matrix.

Once TTC matrices are computed, the following steps are taken:
• The weighted average 1-year probability of default for each rating and for each time horizon is computed by multiplying each TTC
matrix (for the different time horizons) by the PD vector over one year;
• Then the weighted average 1-year PDs are mapped to the closest corresponding ratings, which yields the average degradation
matrix in terms of ratings;
• A downgrade is considered significant, with subsequent classification into Stage 2, if the current rating is worse than the relative
threshold for the respective initial rating and time passed since initial recognition.
Forward-looking information incorporated in the ECL models
The TTC PDs are transformed into PIT PDs by taking in account the macroeconomic environment through a set of macroeconomic
variables, such as real GDP growth rate, inflation rate, unemployment rate, investments, import change , current account balance,
debt, export change etc. This data is sourced from the IMF’s World Economic Outlook, including historical data starting from 1990 up
to today and projections for the upcoming 4 years under three scenarios: baseline, best and adverse. The PIT PD model consists of
a simplified form of the Merton model. In this framework, a systemic variable “ ” which represents the macroeconomic environment
is introduced. The sensitivity of each rating’s PD to this variable is obtained via the calibration of the “ ” parameter. The model takes in
account the default rate of each year and calculates “ ” for each year, which is then regressed with different combinations of macro-
economic variables. Using projections of the macroeconomic variables, the regression formula is used to deduce projections of “ ”,
and based on the one factor Merton model the PIT PDs are obtained. The second PD model considers the default rate per rating in
each year, which enables the calibration of the sensitivity ρi for all ratings.

Measuring ECL – Explanation of inputs, assumptions and estimation techniques
The ECL calculation is based on the following key parameters:
1. Probability of default (PD);
2. Loss given default (LGD); and
3. Exposure at default (EAD).

ECL is estimated under Baseline (typical), Best (favourable) and Adverse (unfavourable) conditions.
The only Point-in-Time estimates are for Probability of Default. LGD assumes Basel estimates and EAD uses amortisation type payment
schedules. Once all components (PD, EAD, LGD) have been computed, the ECL is estimated under three different scenarios: Baseline
(typical), Best (favourable) and Adverse (unfavourable) condition, with weights 52%, 18% and 30% respectively. The final ECL is the
probability-weighted ECL under those three scenarios.