Sovereign Debt Management Forum
October 29-31, 2012
Developing Analytical Models for Public Debt
Management:
Notes from Turkish Experience
Emre BALIBEK, PhD.
Deputy Director General of Public Finance
Turkish Treasury
1
Outline


Why an in-house Model?
Deterministic vs Stochastic Models




Deterministic Model: Stress Testing and
Scenario Analyses
Stochastic Model: Turkish Debt Simulation
Model (TDSM)
Use of Models in Decision Making
Resources and Challenges in Model
Development
2
Why an in-house Model?
3
Why an in-house Model?
An off-the-shelf model (commercial software)



Theoretically provides
 Professionalism
 Efficiency in model implementation
Requires low internal support
May prove to be
 Less flexible
 Restrictive in maintenance and development
 More expensive (in general)
4
Why an in-house Model?
An in-house model
 Requires institutional capacity




Programming skills
Software platform
IT support
Provides



Customized solutions (particularly important
for a developing country)
Independence in terms of development and
maintenance
Lower cost (in general)
5
Why an In-house Model?

A choice between off-the-shelf and in-house
models depends on


User specific needs

Institutional capacity

Cost concerns
Turkish Treasury chose to develop an in-house
model because of its need for flexibility and
sophisticated institutional capacity
6
Deterministic vs Stochastic Models
7
Deterministic vs Stochastic Models

Deterministic Models





Simple
Limited number of scenarios
May not able to sufficently capture the dynamics of
the ‘system’
Easy to build, interpret, and communicate
Stochastic Models




Complex
Better replication of the ‘system’
Hundreds/thousands scenarios
Harder to build, interpret, and communicate
8
Deterministic vs Stochastic Models




Not necessarily substitutes for each other
Approaches similar in essence
Choice should be based on

Available resources

Degree of detail needed

Other country specific circumstances
It is also possible to employ deterministic and
stochastic models as complementary tools

Turkish Case
9
Deterministic Model: Spreadsheet Model


Simple MS Excel-based model
Used to perform
 Scenario analyses and stress tests:




Financing requirement and debt stock projections
under the baseline scenario
Effects of changes in macro-fiscal variables
Scenarios are built by means of expert judgement,
market analysis etc.
Accounting approach is also used for debt
accumulation
10
Stochastic Model: Turkish Debt Simulation
Model (TDSM)
Objective


Assess the sensitivity of public debt to market
movements
Help quantify the costs and risks associated with
alternative financing strategies:
 Provide assistance in developing the strategic
guidelines
11
Stochastic Model: Turkish Debt Simulation
Model
Cost-at-Risk Simulation Framework
Macroeconomic
Scenario
Generation
Expected
cost & risk
of alternative
strategy
Debt
Database
Alternative
Strategies
Cash Flow
Modelling and
Borrowing
Requirement
12
Stochastic Model: Turkish Debt Simulation
Model
Model Building: The Conceptual Model





Decision Horizon: 5 years
Granularity: quarterly
Instruments: A representative selection
Choice of Cost and Risk Metrics
Choice of Debt Structure Metrics (Key Portfolio
Indicators)
13
Stochastic Approach: Turkish Debt
Simulation Model
TDSM v.1 (2003)

Cost Metric: Accrued interest on debt plus changes
in debt amortization due to FX rate movements

Risk Metric: Cost-at-risk (C@R) at chosen
confidence level

Platform: MS Visual Basic and Commercial
Statistical Softwares
14
Stochastic Approach: Turkish Debt
Simulation Model
TDSM v.2 (2007)

Cost Metrics:

Cash-based interest expenditures

Level of debt stock

Risk Metrics: Conditional cost-at-risk (C@R)

Platform: Matlab
15
Stochastic Approach: Turkish Debt
Simulation Model
TDSM v.2.1 (2010)

Modifications based on changing market
conditions and instrument set

Cost Metrics:



Cash-based interest expenditures
Level of debt stock
Level of inflation adjusted debt stock

Risk Metrics: Conditional cost-at-risk (C@R)

Platform: Matlab
16
Use of Model in Decision Making
17
Use of Models in Decision Making


TDSM results are used to determine strategic
benchmarks for debt and risk management
Strategic bechmarks aimed at


Composition of financing (before 2007)
Composition of the debt stock (after 2007)




Allows for a holistic appoach to cover
Outright sales
Buy-backs and debt exchanges
Derivative instruments
18
Use of Models in Decision Making
Illustrative Results
Distribution of Interest Payments
Distribution of Debt Stock Projections
300
90
80
Stock/GDP
Frequency
250
Borç Stoku/GSMH (%)
Frekans
200
CC@R
150
100
50
0
42
70
60
50
40
30
44
46
48
Milyar YTL
Billion TRL
50
52
54
20
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
Years
19
Use of Models in Decision Making
Illustrative Results
Accrued Inflation Adjusted Stock (AIAS) for
Alternative Strategies
Str-2
Str-1
Str-3
Str-5
Str-2
Str-6
Str-4
AIAS @ Risk
Interest Payments @ Risk
Interest Payments for Alternative
Strategies
Str-4
Str-1
Str-6
Str-5
Str-3
Interest Payments
AIAS
20
Use of Models in Decision Making
Strategic Benchmarks

Reduce liquidity / refinancing risk:




Reduce currency risk:


Maintain a certain level reserve of cash
Increase average maturity to the extent that market conditions allow
Decrease the share of debt maturing within 12 months
Borrow mainly in TL
Reduce interest rate risk:


Use fixed rate instruments as the major source of domestic borrowing
Decrease the share of debt which has interest rate re-fixing period
less than 12 months
21
Resources and Challenges in Model
Development
22
Resources and Challenges in Model
Development
Resources Needed



Institutional Capacity

Committed and Skilled Staff

IT systems
Financial Resources

Training

Consulting

Software
Management Support (probably the most important one)
23
Resources and Challenges in Model
Development
Challenges
 In-house modeling

Choice of modelling platform
 Ease of implemention vs.efficiency and speed (Excel vs. some
technical computing platform)
 Capacity building
 Maintenance
 Input Modeling

Lack of long data series
 Stationarity problems in data
 Regime changes
 Financial crises
 Distributional assumptions
 Do scenario generation options enough to cope with extreme
cases (event risk)?
24
For more information:
Turkish Treasury Simulation Model for Debt Strategy Analysis, World Bank Policy
Research Working Paper 6091
25