2 REVISIÓN BIBLIOGRÁFICA
2.1 Regulación de la respuesta del sistema inmune: papel instructivo del sistema inmune
2.2.1 Neisseria meningitidis y sus componentes: capacidad adyuvante
Man’s strength has always been his ability to change his environment to suit himself. He observes the ways things work, and controls disturbances to maintain consistent results, or experiments to improve them. In some instances the risks may be life threatening, but they can be lessened with proper controls. The greater the danger, the greater the need for protection! Business enterprises operate similarly, except the mechanisms involved are more formalised. Control can be viewed as having the following dimensions:
Policies—Rule sets that define qualifying criteria, lending limits of underwriters, risk
mitigation requirements, and so on.
Procedures—Series of actions that must be performed: (i) in order to circumvent policies in
certain circumstances; or (ii) to mitigate risks once some predefined event has occurred.
Structure—Organisational design relating to the level of centralisation, staff roles and
responsibilities, and levels and delegation of authority.
Infrastructure—Resources used for information gathering, computing power, communica-
tions, and deployment.
In extremely simplistic terms, the concepts of structure, infrastructure, and procedures could be likened to management, hardware, and software respectively. Management sets the direction, the hardware provides an engine room, and the software is the grease that ensures everything operates efficiently.
A construct often used to describe the ongoing risk-control process is the ‘feedback loop’, which in more recent years has been changed to the ‘adaptive-control system’. The concepts originated in electronics and engineering, respectively, and are very similar, except the latter is more sophisticated, and tailored to industrial environments. Both refer to mechanisms meant to maintain consistent results, by adjusting inputs to counter changes in output. In its simplest form, the feedback loop has four parts:
Monitor—Ensure that things are going according to plan.
Feedback—Communicate any problems that have been identified. Identify—Determine the source of the problem.
Control—Decide upon and implement a course of action.
The controls involve strategies that fall under four broad headings:
Ignore—Do nothing. This is the easiest and cheapest option, but does nothing to address
Track—Require more information, and give greater scrutiny.
Reduce—Take corrective actions to mitigate the risk, while allowing the process to continue
running.
Eliminate—Get rid of the risk, by shutting down the process to prevent an even worse
situation. It involves high costs/losses, especially when investments are sold at fire-sale prices.
The choice of strategy will be based upon the combination of probability and potential impact. This is illustrated in Figure 5.1, which also provides the equivalents for new business. Other considerations are the ability to respond and cost of response, versus the probability that the action will have the desired effect.
The control process is reactive, and requires a lot of human input when dealing with rare but severe events, such as exogenous shocks (natural disasters, fire, system failure). Disaster recov- ery plans can be implemented, but just how they are implemented will vary depending upon the situation. In contrast, if the risks are known and recurring, it is possible to set up a pro- active risk management system. The controls are designed and implemented as need requires, which is usually the case for retail credit.
5.1
Adaptive control
The adaptive-control system (see Figure 5.2)1is a further evolution of the feedback loop. The
concept has been hijacked from automated environments where closed systems are used to 110 Module B : Risky business
High Low High Low Probability Reduce Cost Impact Track Ignore Eliminate New business Existing business Avoid Seek Opportunity Accept Decline
Figure 5.1. Risk strategies.
manage highly technical processes, like Boeing 737 autopilots, and NASA robots. Adaptive- control systems have four main building blocks:
Process—The business function being controlled, whether application processing, account
management, collections, marketing, etc.
Controller—Governs the process, by executing a set of parameterised instructions.
Identifier—Measures consistency of operation, reports the source and magnitude of any dis-
tortions, and where significant, motivates changes.
Design—Determines the changes needed to address the distortions.
A conventional control system only has the first two components, being the controller and the process. It is the last two, identification and design, that are the essential and active ingredients of an ‘adaptive-control’ system. Within this model, there are also flows of information:
Reference input—Parameters that set the framework.
Control signal—Communicates parameters from the controller to the process.
Output—Results generated by the process, such as bad debts, attrition, number of customers,
revenue, and so on.
Disturbances—Distortions that arise both inside and outside the process, which are identi-
fied by strict monitoring and comparison against expectations.
Parameter changes—Design changes to controller parameters.
Controller changes—Modifications to the design of the controller itself, like adding new
information sources and capabilities.
Consumer credit control systems such as Probe™, TRIAD™, Falcon™, and others use a sim- ilar approach, but there is still a lot of human intervention to adjust strategies (controller). Closed systems are as of yet infeasible for credit decision-making.
In general, this is a reactive approach to risk management, which is appropriate for man- aging existing products and markets. If the economy is slowing and people are starting to default, then it is wise to be stricter in collections. If competitors are stealing customers, or the
Output Plant/Process Design Controller Reference input Parameter changes Control signal Identifier Disturbances Model Controller changes Figure 5.2. Adaptive-control process.
demand for credit is reducing, then a change in cut-off strategy may be in order. If, on the other hand, the lender wants to be proactive and take on more or less risk, or risks of different types, then changes to the reference inputs are required. There are, however, extra risks that come with fiddling with the system. The best way to determine whether a proposed change will have the desired effect is to apply some scientific rigour, which is covered next.
5.2
Be the master, not the slave
If you want to make enemies, try to change something. Woodrow Wilson, 28th US president, 1856–1924
In the twenty-first century, people have to deal with ever-increasing amounts of change, both as observer and participant, driver and passenger. Credit scoring was initially developed to model a world that was still fairly steady; men had lifetime employment, women stayed at home, and the kids rushed home to watch the Flintstones at 4:30 pm. That was not enough though— somebody had to bring in the wooden spoon and start stirring things up, with ‘What if?’
This question can only be answered if there are tools in place to aid conscious and calculated changes, while ensuring that the proper risk/return dynamic is maintained. Within the realm of business, the term ‘science’ is commonly used. Arsham (2002) refers to the decision sciences as those commonly known as management science, success science, and operations research. But why use the word ‘science’? For many, it is a much loathed subject from their school days, and if they did not hate it, the answer to the above question might make them hate it! The follow- ing is a summary of ideas presented by Bala (2001), which is further summarised in Table 5.1.
A brief history of scientific philosophy—an essay
Mankind has spent much of his existence trying to understand and control his environ- ment. True knowledge was gained haphazardly, while myths and legends were used to explain the rest. The first to use a structured approach was Euclid, the fifth-century BC
Greek, who developed geometry using an axiomatic system based on infallible truths. Unfortunately, this only works in mathematics, and fails when trying to learn about nature. It was more than a thousand years after the fall of the Roman Empire before seventeenth- century Renaissance thinkers questioned the dogma of the church. The findings of Galileo and others caused philosophers to search for fail-safe frameworks for discovery. Francis Bacon (1561–1626) proposed deductive reasoning, based on observation and experimen- tation, what we know as the ‘scientific method’. René Descartes (1596–1650), of ‘I think, therefore I am’ fame, proposed inductive reasoning based on mathematics and pure logic— a deconstructive approach of breaking a problem down into its parts. English and French scientists were divided along these Baconian empiricist and Cartesian rationalist lines for a century. During this period, the concept of science as a structured and continuing search for knowledge arose; a search that uses hypotheses, theories, laws, concepts, and models as tools of discovery.
Neither the empiricist nor rationalist approach worked when Isaac Newton (1642–1727) tried to explain gravity. He instead used a synthesis of both to come up with his mechanis- tic view of the universe. The Newtonian normativist view relies on expectations that scientific methods will produce principles that accurately predict and explain a variety of phenomena. Theories were measured by explanatory adequacy, predictive accuracy, scope of success, simplicity of assumptions, etc.
All pre-twentieth-century scientific thought was directed at mapping a mechanistic uni- verse, but the map had some wrong turns. In 1905, Albert Einstein (1879–1955) published his special theory of relativity, which showed that many Newtonian rules become invalid as the speed of light is neared, and in 1927, Werner Heisenberg (1901–1976) came up with quantum mechanics, which ruled out certainty and impossibility at the sub-atomic level. This is not to say that these newer models are truths; they are simply the best representa- tions that we currently have. Scientists are still trying to come up with a grand unifying theory, to marry the theories of the very big (relativity) and very small (quantum). One pos- sibility is string theory, which describes everything as infinitesimally small threads, but it is not quite there yet.
Einstein’s and Heisenberg’s theories contested all knowledge of physics derived using Newtonian normativism, which forced scientists to look out from under their determinist umbrella, and revise much of what was previously thought sacrosanct. They upset the idea that slow accretion of knowledge, using accepted frameworks, was a fail-safe way of devel- oping scientific knowledge. The Newtonian approach is not invalid though; it is a special case within Einstein’s bigger and richer description. Experience, reason, and norms still form the basis of most knowledge discovery today.
The scientific philosophies developed by Bacon, Descartes, and Newton are applied mostly to the hard sciences; the natural sciences that we usually associate with the earth and the stars—like physics, chemistry, biology—which can be subjected to the rigours of the scientific method, and the results relied upon for centuries. Since the early twentieth century, these philosophies have also been increasingly applied to the soft sciences; the social sciences that deal with man’s relationship with both man and his environment—like psychology, sociology, economics, etc. These relationships are fluid though, and scientifi- cally derived results have shorter shelf lives. Both man and society change, and explana- tions will change along with them. Within the soft sciences, and even most hard sciences outside of physics, the normative approach should be viewed as sufficient, but any theories must recognise possible changes to the assumptions upon which they were based.
Table 5.1. Philosophies of science
Francis Bacon René Descartes Isaac Newton
Baconian empiricism Cartesian rationalism Newtonian normativism
Observation/experimentation Logic/reasoning Expectations to qualify
Deductive Inductive Synthesis
Scientific principles were not used in business until the early twentieth century, when they were popularised by Frederick W. Taylor’s 1911 ‘Principles of Scientific Management’. Taylor focused on reducing the amount of time taken for production-line tasks, with the sometimes inhuman use of a stopwatch (Arsham 2002). The field of operations research only came into being after the Second World War, to develop new methods of dealing with complex logistical problems, especially those of managing huge armies and ensuring they are well supplied. The advent of computers then allowed these same tools to be applied increasingly in business. [Management Science/Operations Research is] a scientific method of providing executive management with a quantitative base for decisions regarding operations under their control.
Mores-Kimball
Arsham (2002) also highlights that decision sciences differ from other sciences, in that there is a decision-maker, who usually is not—and should not be—one and the same per- son as the analyst/scientist. Analysts thus require communication skills to contextualise the results and express them in terms that laymen can understand.
Credit scoring and the scientific method
What does that have to do with credit scoring? Credit scoring is a tool used for credit decision- ing, a decision science that falls within the realm of economics, which in turn falls within the social sciences. Hence, some of the frameworks used in science can be borrowed, but organisa- tions must be able to modify not only their assumptions, but also strategies and resources, going forward. Destination is more important than journey however, and lenders are more interested in ‘what’ will happen than ‘why’ (although the latter is definitely a bonus).
The scientific method is usually presented as a process of experimentation involving:
Observe—Viewing and describing the world around us.
Hypothesise—An attempt at explanation of either the observed or related phenomena. Experiment—Use the hypothesis as a prediction tool.
Decide—Use the results of the experiment to accept or reject the hypothesis.
If the experiment’s results support the hypothesis, it is used as the basis for a theory or law, oth- erwise it is rejected and the researcher tries again. The goal is to come up with a hypothesis that satisfactorily predicts the outcome of the experiment (Wolfs 2002). This is the empiricist 114 Module B : Risky business
Table 5.2. Experimental design frameworks
Scientific method Bisgaard et al. (2002) Arsham (2002) Generic
Observe Define Perceive Design
Hypothesise Measure Explore Execute
Experiment Analyse Predict Analyse
Decide Improve Select Improve
approach—the only things missing are the deconstructive and repetitive aspects. If the hypoth- esis is rejected, or even if it is accepted, the problem can be broken down into further parts for greater understanding.
This text’s interest is in how the scientific method is used in the retail credit environment. While there have been no frameworks presented specifically for that context, there are several for business applications, which have marked similarities (Table 5.2). In all instances, the goal is to provide a formalised approach for learning and decision-making:
Bisgaard et al. (2002)
Define—The problem statement, presented in measurable and actionable terms. Measure—Observe variation of performance data from that expected.
Analyse—Determine the sources of variation, or reasons for success/failure. Improve—Modify process drivers to achieve desired goals.
Control—Hold on to the gains.
Bisgaard’s framework is the same as that used as the backbone of Six Sigma, which is a scientific means of achieving continual improvement in business processes, by focusing upon quality and service issues to reduce the costs of poor quality and customer dissatis- faction. See Keller (2005).
Arsham (2002)
Perceive—Realise that there is a problem, a need, or a goal to be achieved. Explore—Find out the set of possible actions that can be taken.
Predict—Try to determine the outcome for each of the identified alternatives. Select—Choose the best alternative, based on both effectiveness and risk. Implement—Put the chosen action into place.
General experimental design
Design—Plan the experiment: define the problem, objectives, and possible solution(s);
identify the input and output variables, and the quantity and type of data needed.
Execute—Perform the experiment: select the sample, apply possible solution(s), and collect
the data.
Analyse—Analyse the data against all of the output measures. These will include measures
relating to cost of the action, and effectiveness at achieving the desired result.
Improve—Determine whether the (or which) challenger strategy is good enough to replace
the champion, communicate the results, and motivate to have it implemented.
As indicated earlier, the primary defining feature of the soft sciences and business is how quickly the assumptions change. For the latter, this applies especially in times of rapid growth,
when both organisations and their processes evolve through repeated cycles of increasing complexity and simplification to achieve optimal performance.
5.2.1 Champion/challenger
The usual approach to business is very reactive, if only because so much time is spent fighting fires that management and staff members have little energy to be proactive. The problem is, however, that in today’s rapidly changing world, companies have to be the architects of change, not the victims. Lenders should continuously be considering questions like, ‘What if we phone instead of mail?’, ‘What if we charge more for NSF cheques?’, and ‘What if we provide higher limits?’.
Adaptive-control systems in the credit scoring world also have a feature called champion/ challenger, a ‘what if?’ tool that allows performance comparisons of the new and novel against the tried and tested, so that lenders can have a forward view, instead of always looking in the rear-view mirror. The champion strategy is always the dominant one currently in place, which has worked in the past, and is trusted. The challenger strategy is the underdog—the contender that has to prove itself before gaining acceptance. Only if the challenger wins can it become the new champion.
According to Thomas et al. (2002) the champion/challenger approach is also commonly used: (i) in medicine, for testing drugs and treatments for various ailments; and (ii) by com- panies looking to introduce new versions of a product, such as toothpaste, to the market.
The use of a champion/challenger approach limits the risks that might arise from hasty implementation of poorly thought out strategies. According to Thomas et al. (2002:164), it can be used at any stage in the risk management process, as long as it is possible to: (i) identify a random sample of cases, say 5 per cent, where the challenger can be applied; and (ii) track their performance separately. The final decision on whether to accept the challenger will then take into consideration:
Marginal cost or benefit—of using the new strategy.
Effectiveness—of providing the desired results, in terms of risk (defaults, losses), value
(assets, revenue), or attrition (dormancies, closures).
Positioning—whether it has caused customer complaints, or may impact on the lender’s
reputation.
While experimentation is an extremely powerful tool, there are limitations: (i) the number of challenger cases is often too small to draw proper conclusions; (ii) it takes time before the chal- lenger’s effect on the test group becomes apparent; and (iii) there is always the danger that even though the experiment yielded positive results, the challenger might be disastrous when imple- mented in full—especially if the business environment changes significantly.
These limitations could, at least partially, be countered by using simulation. This is a process often used in the sciences, which has come to be one of the recent buzzwords in the credit