Lending to Small and Medium Enterprises:
A Novel Approach to Credit Portfolio Management
Andras K6vari — a.kovari@student.tudelft.nl — Delft University of Technology, The Netherlands
Erik Pruyt — e.pruyt@tudelft.nl — Delft University of Technology, The Netherlands
Abstract
There is a vast unmet demand for credit from SMEs in emerging markets, while banks need novel ways of
approaching their risk strategies. These inspired a pilot project within a major commercial bank. The goal of the project is to
investigate in what ways exploratory system dynamics (SD) could support lending decisions and monitoring of a SME credit
portfolio. This paper reports on this project as a practical application of SD modeling for a bank. The first stage of the
project consists of ‘quick and dirty’ modeling and analysis intended to illustrate the possibilities of this approach. A generic
model of a company with debts was modeled to explore plausible dynamics given the assumptions about how the
company’s envi may affect its p ‘e. The 2” and 3” stage — modeling of actual companies and aggregating
them into a portfolio for stress testing — is work in progress due before the summer.
Keywords: SME credit, portfolio risk management, System Dynamics, Exploratory Modelling & Analysis
1. Introduction
Small and Medium-sized Enterprises (SMEs) are the backbone of growth and employment in
emerging economies. They represent 95% of all companies, and employ two thirds of the labor
force. A report by McKinsey & Co. estimated that SMEs in emerging markets (EM) account only for
7% of the total value of loans, bonds and public equity outstanding (Stein, Goland, & Schiff, 2010).
Despite their huge share in job creation, SMEs’ capital employed in some form of credit is just a
fraction of the loans that the large corporations use. This is found by some to be the biggest
constraint on these companies’ growth (Beck & Demirguc-Kunt, 2006).
Lending to SMEs is riskier, but the yields that investors can get from larger companies have
long reached a plateau. There is a huge unmet credit demand from the SME side. Therefore, banks
that can develop a sound business model for lending to them could find great alternatives for future
revenue streams while contributing to economic development and job creation. However,
confidence in banks’ sound business models and risk management in general has hardly ever been
lower. Bankruptcies and bail-outs during the crisis to high-profile scandals’, all threatened to
undermine bank’s credibility in proper risk management. Angered economists also talk about
financialization of industry, warn about debt deflation and call for the restoration of banks’ role as a
service to industry (Hudson, 2012). KPMG reports on the expectations of risk management being
outpaced by capabilities (KPMG and The Economist, 2013), while another survey calls for the
rethinking of risk strategies within banks (Ernst & Young, 2010). The European Central Bank stated
that there is an enormous need for further research into the management of financial systems and
systemic risks (ECB, 2010). Novel analytical tools aiding banks’ decision-making could help restoring
confidence and reconcile finance and industry.
* Think of insider trading, HSBC money-laundering, Barclay’s Libor fixing, JP Morgan’s recent 6bn losses
Lending to SMEs | 2013
Banks’ need for novel ways of risk approaches and a vast unmet demand for credit from
SMEs in emerging markets inspired a pilot project within a major commercial bank. This paper
reports on this project as a practical application of SD modeling. A small group within this bank’ is
considering alternative ways to extend credit to SMEs in one of the booming countries of the
emerging markets, and wants to combine this with innovation in its approach to uncertainty and
risk. The goal of the project is to investigate in what ways SD modeling combined with an
exploratory approach could support lending decisions and monitoring of an SME credit portfolio.
The rest of this paper is organized as follows. In section 2 we start with contrasting the
current practices of lending decisions with the approach envisioned in this project. Then we mention
some of the background literature that supports our work. Section 3 illustrates the exploratory
analysis method using a simple SD model’. Finally Section 4 looks ahead at what still needs to be
done to further develop the project.
2. A Novel Approach?
Access to finance by SMEs is low for various reasons, but a lack of credit history and detailed
archives of performance documentation is a major barrier. When an SME does get into the
negotiations, an examination of the cash flow perspectives follows usually on spreadsheets. The
models built are then usually linear and contain expectations about the future values of parameters
and coefficients. There are also assumptions about how these parameters might change in the
future. However, the resulting forecasts can be as good as the assumptions that generate them.
Modifying these assumptions one-by-one can give a sense of sensitivity, but there is hardly a
systematic evaluation of possible future scenarios. Assessing all the uncertainty space and then
deriving meaningful insights is virtually impossible. Such an attempt would quickly run into the
limitations of linear modeling and multiple regression analyses that provide some of the coefficients.
Performance of a company through time is driven by delays and regenerative feedback loops
in its processes and interactions with its environment. Current modeling approaches do not reflect
these features, therefore they can be misleading especially when a broad range of uncertainties
must be evaluated. These limitations were already documented by Forrester in his early years of
studying industrial models and applying his control systems background. A seminar note from 1956
published almost 50 years later contains the original ideas (Jay W. Forrester, 2003). SD modeling can
capture the feedback loops of a dynamic system and its stock-flow approach is well suited for
representing company operations and the developments of a balance sheet.
The plan of approach for this project is illustrated in Figure 1. High-level SD models of the
companies’ main operations are built for each of the SME that constitute a portfolio. The models
should capture the main drivers behind the performance of the company. Some of these factors will
be external variables that are not under the direct control of the company (or the creditor for that
matter). Examples are availability of a key resource, demographics of the client base or the stability
of a major supplier. These form the immediate environment in which the company operates, and
they are in turn influenced by shifts in the macroeconomic level. In what way and to what extent
? References to the details of the project will remain vague due to confidentiality
3 Acknowledgements to Santiago Braje for inspiration on this section
2Den
Lending to SMEs | 2013
these factors influence the other is uncertain. Exchange rates, oil price and emigration can have
direct or indirect effect on a company’s performance, and this effect can also take different forms
from strongly positive through mixed to very negative.
e Macroeconomic factors
Industry-level / regional
4 factors
O po
—
>]
\ SME Credit Portfolio
XN
Figure 1: Conceptual model of the portfolio stress testing bench
An exploratory approach tries to consider all the possible hypotheses about how factors in
the company’s environment influence its performance. These hypotheses are then turned into
alternative ‘models of the future’. Combining the plausible ranges for the external parameters and
switching between alternative structures a large number of plausible future scenarios can be
generated. The portfolio then can be ‘stress-tested’ for extreme circumstances or interesting
combinations of events.
A similar methodological approach is called Exploratory Modeling and Analysis (EMA) and
was extensively described by (Agusdinata, 2008; Lempert, Popper, & Bankes, 2003). EMA treats
questions like: “what is the range of plausible future dynamic devel of a phi of
interest? Under what circumstances can we expect which dynamic developments?” (Jan H. Kwakkel &
Pruyt, 2013) Treating such questions would support the risk approach towards volatile SMEs that can
be very much influenced by events in their operating environment. Analysis tools that can help
answering these questions are readily available open source’ and a growing number of works
demonstrate their use (Auping, Pruyt, & Kwakkel, 2012; Hamarat, Kwakkel, & Pruyt, 2013; K6vari &
Pruyt, 2012; J H Kwakkel & Timmermans, 2012)
This modelling project involves building small models of company operations and industry
dynamics. This effort is supported by a vast SD literature from (Jay Wright Forrester, 1961) through
(Sterman, 2000) to (Warren, 2008). There are also some papers that deal with bank-related
modelling with SD. Some of them describe banking crises, bank runs and build models that could (or
could have been) used for crisis management (Pruyt & Hamarat, 2010; Pruyt, 2009; Rafferty, 2008).
However, there are also a few published SD works related to risk management of a commercial bank
(Chaim & Castellano, 2012) and stress testing of a financial system from the point of view of a
central bank (Anderson et al., 2011). (Gramlich & Oet, 2012) give a comprehensive overview of the
modelling works related to systemic financial feedbacks (SFFs). They find that SD is a suitable
* For example R packages or the EMA workbench at http://simulation.tbm.tudelft.nl/
-3-
Lending to SMEs | 2013
modelling approach for such systems and problems. However they note that “The scarcity of SD
modelling for SFFs may be attributed to the lack of required economically-sound foundations for
theoretical modelling”. An exploratory approach circumvents this problem by looking at all plausible
model versions rather than coming up with the model of the true theory.
3. Asimple SD model for illustration
In this section we illustrate the above presented approach with a simple SD model of a company that
tries to grow while initially taking on a loan.
3.1 The Model
At this high level, the business is seen as a circular flow of money that tries to create profits by
constantly investing its available money into assets. Assets are then partly monetized (sold), thereby
generating revenues, possibly at a margin. Assets that are not monetized usually depreciate. There is
also a debt that the business takes or considers to take. Traditionally an amount of loan is given that
bears interest. Experience shows that “it is never the right time” for companies to repay part of their
debt, mostly only paying the minimum required by their contract, or taking on new loans to pay
(only) the interest on an old loan. Therefore they face uncertainty in terms of what interest rate they
can secure.
Three additional hypotheses are proposed (illustrated on Figure2):
e The bigger the company becomes the lower monetization rate it can achieve on its assets.
e As the company grows bigger, the fluctuations in the monetization rate are reduced
e As the company grows, the interest rates at which loans can be refinanced also tend to reduce
— -
sales“
+
: f \
( \ \
nv @ Ch af } fluctuation in
average monetization rate
; monetization rate
+_assets _
7 \ > Ps
sustainability } - -
interest rate
debt ff ~*~
Figure 2: Conceptual model of a growing company with a loan
The stock-flow diagram of the model is shown in Figure 3. For illustration the initial data can
be arbitrary, Cash and Debt are initialized at 200 and 100 respectively, while Assets Book Value is 1
only to avoid division by zero at the outset. The business starts by investing all its money into assets,
meaning that the investment rate is 100%. Selling assets will generate sales with a so called
monetization rate, while this will create an inflow of money possibly at a margin. An average delay in
receivables is added, to examine the effect of late payments. The parameters of the model are
summarized and explained in Table 1.
Lending to SMEs | 2013
uone
roaadap:
<So! XO:
\ cennanars
TYseozps}asse
B]0}>——— Jo mols aaneyar
SJass
SEO
air
‘ae -
<dals
. sesuadxa we queurkedes 3p
Sd Sy S 4 warRET
ee | (cra _
a]pdo sseursng q ra oneA FOO SISSY] yoyEDardap se) ica
\ | ‘ayer monezt
a” paisnipe azis J of Seen
<eU[>—e — ToHeMony anuaaar . ‘osvOroU 199}
= ® sayes aie uoneisaidep JBA0 ep 1qepI /
Pa ayer uoHezauou /
Se Appuow /
guoyenionyy ee A ati ayes ysalaqut ayer ysereyur
aywanoe E “*—paysnlpe azis
ne so]QeAteoor UL
quonanpe! ayer woneznonont Azjep azeroan |
CO SESS aeione ayeanoe sae cojep i
Posy
moss oayeyjer
aut] Arpiqeureysns
sutures JeA0 199
Yseo:
<iqeq>
A
yoog sjassy:
a _e
sjasse ]B}0}
a
Aynba
ysouoqU poxty
Figure 3. Stock-Flow Diagram of the simulation model with some possible indicators at the bottom
Lending to SMEs | 2013
Uncertainty Description Range
activate interest reduction? | Will the interest rate decrease with the growth of the company? y/n
fixed interest rate If size-related reduction is deactivated, a fixed interest rate can be used | 0-50%
interest adjustment factor _| A parameter setting how fast will interest rate decrease 1-100
activate average reduction? | Will the average monetization rate reduce as the company grows? y/n
monetization adjustment f _| A parameter setting how fast will the monetization reduce 1-100
fixed monetization rate If feedback reduction is deactivated, a fixed rate can be used 0-100%
time to invest the time it takes for assets to be acquired 1-6 months
investment rate share of cash is invested into assets instead of debt repayment 80-100%
average delay in receivables | delay in the actual inflow of money from sales 0,2-6 months
depreciation rate How fast will the unsold assets depreciate? 0-100%
margin Uncertainty in the margins that the company can reach 1-1000%
amplitude adjustment f A parameter setting how fast will fluctuations reduce 1-100
activate fluctuation? Will fluctuation reduce as the company is growing? y/n
business cycle length From yearly to decade long fluctuation - depending on the case 1-10years
present cycle position O=middle of upturn; n/2=top of the wave; n=middle of downturn; 0-1,5-3,1-4,7
3n/2=depth of crisis-facing upturn
Table 1: External parameters and their range considered in the analysis
The switch -parameters with question marks can be used to deactivate their respective
loops. This can help assessing the impact of the three hypotheses mentioned above. Each of these
hypotheses takes the Assets Book Value as a proxy for the size of the company and influences the
fluctuation, interest rate and monthly monetization rate. However, there is uncertainty about how
quickly the growth of the company can be felt on these rates. Therefore an adjustment factor is used
to vary the actual effect (see Figure 5). For example: size adjusted interest rate
= adjustment factor/Asset Book Value. Table 1 also contains suggestions for the ranges of each
parameter that can be analyzed. Some of them represent what is physically possible (like
depreciation rate), others are just suggested plausible ranges (like margin).
monetization rate2
1
0.75
0.5
0 30 60 90 120
Figure 5. Varying the adjustment factors can change speed at which monetization rate decreases due to growth.
Fluctuations in the sales can be modeled to represent long-term business cycles, or seasonal
variations. Some well-known key performance indicators were modeled to track the changes over
time. The sustainability line is the difference between the relative growth in the company’s assets
and cash and the relative growth of debt. Above zero on this indicator means that on the long term
the company is on a sustainable track. The model could be further decorated with many
assumptions as one wishes, similarly to the above presented ones. This selection of features and
hypotheses is plausible, but is not meant to represent ‘the most important things’ that can happen
to a company. They are selected somewhat arbitrarily for illustration.
Lending to SMEs | 2013
Figures 5 and 6 show the result of four simulation runs. The first is the result of fixed rates
with all the 3 hypotheses turned off. Since depreciation is relatively low, the company goes on an
exponential growth path. Once the balancing effect is turned on in the 2™ run, the growth becomes
limited.
Assets Book Value monthly monetization rate
1
0.75
05
0.25
of FREE
12 24 36 48 60 72 84 96 108 120 i 04364860728 08120
Time (Month) Time (Month)
Assets Book
‘Assets Book
‘Assets Book
‘Assets Book Ve
Figure 5. Si ion results of activating the ‘one by one. The 2" and 3 runs are overlapping
Debt sustainability line
0 12 2 36 48 60 72 84 96 108 120 o2
Time (Month) 1 36 Sf 72 90 108
Debt Time (Month)
Debt: Activa
Debt
Debt
Figure 6. Si ion results of activating the one by one. 1°* and 2 runs are overlapping for Debt
Many pages could be spent continuing with exploring the effects of individual parameters
asking the ‘what if?’ question. Programs such as Vensim that can display the simulation results ‘on
the fly’ as parameters are varied are useful tools to give a good sense of the influence each external
parameter has on the model. However, the aim was a more systematic approach. In what follows,
the full uncertainty ranges mentioned in Table 1 are explored.
3.2 Beyond ‘What if?’ Questions
One question that the commissioner of this project found important is: what can possibly
happen? In some cases the future evolution of external parameters is uncertain, but we can have
information on what is the possible range within which they can vary. Or it might be insightful to see
what are the possible dynamics that can occur in extreme circumstances. Figure 7 presents the
a
Lending to SMEs | 2013
outcomes: hundreds of scenarios generated sampling through the given uncertainty ranges. A blue
envelope shows what the ranges of possibilities are at a glance. Visual inspection of the individual
lines can indicate what kinds of dynamics occur. However, recent techniques such as time series
clustering (Warren Liao, 2005) can make the process automated.
Figure 7. Possible dynamics with a full-range exploration
On the graphs presented in Figure 7 the distribution of the end states for the scenarios are
also displayed. This gives an indication of how many of the scenarios end up in a particular state.
Debt for example will mostly be below 2000 after 10 years, but there are still quite a few runs that
go way above that.
Once we see the range of possibilities, the next logical question is: what combination of
parameters can lead to which scenario? What leads to a debt higher than 5000 after 10 years?
Another analysis tool, the PRIM algorithm can reveal the answer. The algorithm searches among the
uncertainty ranges for combinations of parameters that lead to a given end state. Its answer to the
question of what leads to debt greater than 5000 at the end of the runs is displayed on Figure 8. The
combination of ranges is displayed on normalized scales. The result is not very surprising: investment
rates at the high end (meaning little expenditure to repay debt) are the defining factor. Margins at
the lower end, and slowly decreasing interest rates also contribute. The 5 parameters are ordered
according to the length of the lines displayed. It is a ranking of how crucial that parameter is in
leading to the undesired state.
Lending to SMEs | 2013
investment rate —
margin
interest adjustment factor
monetization adjustment factor
rate
Figure 8. PRIM: the combination of parameter ranges that lead to high Debt
4. Next steps
The above presented model and analysis is a metaphor to a company. It was used to show
how EMA can help exploring the risks involved in a commercial loan. The model used was simple and
abstract, therefore the results of the analysis also somewhat unsurprising. It was used to illustrate
the tools available and to build confidence in the potential of this method to the commissioner. We
have to turn from this abstract model to a more operational representation of the actual companies
and their environment to make the analysis relevant for use in decision-making.
The second stage of this project is to turn to the descriptions of the companies within the
existing portfolio and modeling them. There the challenge is to find a balance between detailed
modeling of a company and simplicity for the sake of efficiency in use. After all the modeling should
not take too much time, but it should be just enough to facilitate a meaningful discussion that adds
real value to the already existing practices. A possible road is aiming for a handful of generic types of
SMEs, and then trying to find ways of quick customization for the individual aspects of the actual
companies that make up a portfolio. Similar efforts are documented in (Winch & Arthur, 2002).
The third stage is then the exciting part: aggregating the models into a portfolio, building
hypotheses about external effects and stress-test the resulting model. Then a debate can follow
about what insights can be gained from the analysis that is relevant for the actual case.
5. Conclusions
An innovative approach towards risk and uncertainty in a bank was presented and illustrated
ona simple SD model. It is a first stage of a work in progress due before the summer. The goal is to
find ways in which the presented exploratory approach can support lending decisions and portfolio
monitoring within the bank. The discussion in this paper already points to potentially valuable tools.
The tools and analysis methods presented in this paper were mostly taken from the EMA
literature: “EMA ... can be used to address ‘beyond what if’ questions...Because of this focus, EMA
-9-
Lending to SMEs | 2013
stimulates ‘out of the box’ thinking and can support the development of adaptive plans or
policies.”(Jan H. Kwakkel & Pruyt, 2013). We find such an approach helpful and desirable for the
evaluation and monitoring of credit portfolios. The portfolio stress-testing workbench developed
might not fully replace existing investment analyses, but it could provide valuable support to build a
robust business model for future lending to SMEs. Further work from this project will be included in
this paper.
As with any exploratory research, the outcome of this project is unknown: will it be worth
doing all the modeling and analysis? Does it really add value to the existing practices? Future
research could try to measure that. Such an inquiry can also help to find analysis methods that do
make an impact and could point to directions in which the EMA toolkit could be extended.
References
Agusdinata, D. B. (2008). Exploratory modeling and analysis: A promising method to deal with deep uncertainty. Delft
University of Technology.
Anderson, S., Long, C., Jansen, C., Affeldt, F., Rust, J. W., & Seas, B. (2011). Dynamically Stress Testing Financial Systems.
Proceedings of the International Conference of the System Dynamics Society.
Auping, W., Pruyt, E., & Kwakkel, J. (2012). Analysing the Uncertain Future of Copper with Three Exploratory System
Dynamics Models. Proceedings of the International Conference of the System Dynamics Society.
Beck, T., & Demirguc-Kunt, A. (2006). Small and medium-size enterprises: Access to finance as a growth constraint. Journal
of Banking & Finance, 30(11), 2931-2943. doi:10.1016/j.jbankfin.2006.05.009
Chaim, R. M., & Castellano, M. (2012). ERM quantitative risk analysis methods and techniques applied to a small
commercial bank. Proceedings of the International Conference of the System Dynamics Society.
ECB. (2010). Towards Macro-Financial Models with Realistic Characterizations of Financial Instability. Financial Stability
Review, (December), 138-146.
Ernst & Young. (2010). Recover , adapt , advance - Back to business in an uncertain world. Retrieved from
http://www.ey.com/Publication/vwLUAssets/Recover, adapt, advance:_back_to_business_in_an_uncertain_world
/SFILE/Rec dapt-advance_back-to-business-i ertain-world.pdf
Forrester, Jay W. (2003). Dynamic models of economic systems and industrial organizations. System Dynamics Review,
19(4), 329-345. doi:10.1002/sdr.284
Forrester, Jay Wright. (1961). Industrial Dynamics , Productivity Pr.
Gramlich, D., & Oet, M. V. (2012). Systemic Financial Feedbacks — Conceptual Framework and Modeling Implications.
Proceedings of the International Conference of the System Dynamics Society.
Hamarat, C., Kwakkel, J. H., & Pruyt, E. (2013). Adaptive Robust Design under deep uncertainty. Technological Forecasting
and Social Change, 80(3), 408-418. doi:10.1016/j.techfore.2012.10.004
Hudson, M. (2012). THE BUBBLE AND BEYOND. ISLET.
Kévari, A., & Pruyt, E. (2012). Prostitution and Human Trafficking : A model-based exploration and policy analysis
Prostitution and Human trafficking 2012. Proceedings of the International Conference of the System Dynamics
Society.
-10-
Lending to SMEs | 2013
KPMG and The ist. (2013). ions of Risk Outpacing Capabilities - It’s Time For Action.
Retrieved from http://www.kpmg.c¢ i ticlesPublicati ris
outpacing- iliti ions-risk- vey.pdf
Kwakkel, J H, & Timmermans, J. S. (2012). Safe Operating Spaces for Human Water Use : Applying Exploratory Modeling
and Patient Rule Induction to ANEMI.
kwakkel, Jan H., & Pruyt, E. (2013). Exploratory Modeling and Analysis, an approach for model-based foresight under deep
uncertainty. Technological Forecasting and Social Change, 80(3), 419-431. doi:10.1016/j.techfore.2012.10.005
Lempert, R. J., Popper, S. W., & Bankes, S. C. (2003). Shaping the Next One Hundred Years: New Methods for Quantitative,
Long-Term Policy Analysis. (RAND, Ed.). Santa Monica, California.
Pruyt, E. (2009). Saving a Bank ? Cracking the Case of the Fortis Bank ? A High-Level System Dynamics Simulation Model.
Proceedings of the International Conference of the System Dynamics Society.
Pruyt, E., & Hamarat, C. (2010). The Concerted Run on the DSB Bank : An Exploratory System Dynamics Approach.
Proceedings of the International Conference of the System Dynamics Society.
Rafferty, M. (2008). Northern Rock plc: A case study in banking policy during times of duress. Proceedings of the
International Conference of the System Dynamics Society.
Stein, P., Goland, T., & Schiff, R. (2010). Two trillion and counting. Retrieved from
http://www.mspartners.org/download/Twotrillion.pdf
Sterman, J. D. (2000). Business Dynamics: Systems Thinking and Modeling for a Complex World. McGraw-Hill/Irwin.
Warren, K. (2008). Strategic Management Dynamics. John Wiley & Sons, Ltd.
Warren Liao, T. (2005). Clustering of time series data—a survey. Pattern Recognition, 38(11), 1857-1874.
doi:10.1016/j.patcog.2005.01.025,
Winch, G. W., & Arthur, D. J. W. (2002). User-parameterised generic models: a solution to the conundrum of modelling
access for SMEs? System Dynamics Review, 18(3), 339-357. doi:10.1002/sdr.252
-11-
