Routines are computer algorithms that, in economic models, simulate choices, activities and events, which in the real world are carried out in routinized and non-routinized ways. In this paper, a method to assess how well the routines correspond to real data is proposed.
Moreover, since economic models are not "simplification of reality" but they are rather "complexification of arguments", the discussion of routines and their combined effects in the models is a gym tool to mature exploration of non-linear possible worlds, allowing for policy "thought experiments" and suggestions, whose actual implementation requires a thorough analysis of the targeted reality.
Routines consist in repetitive and well-structured procedures but they can include one or more random components, with the result of being capable of simulating fairly erratic behaviours.
In their lives, people follow routines to choose and take action, relying on a limited set of pieces of information, assessments, intuition and mood, possibly in communication with others.
People, organizations and interaction spaces behave according to a wide number of routines, activated, replicated, selected, inherited, modified and adapted keeping into account specific internal and external contexts and circumstances.
Routines are executed responding to certain trigger conditions and leveraging parametres and value constellations that reflect contingent situations. Routines have usually roles and goals; however, their effectiveness in performing the former and reaching the latter relies often on other routines being performed by the same agent or by others.
"Most of what is regular and predictable about business behavior is plausibly subsumed under the heading 'routine', especially if we understand that term to include the relatively constant dispositions and strategic heuristics that shape the approach of a firm to the nonroutine problems it faces. The fact that not all business behavior follows regular and predictable patterns is accommodated in evolutionary theory by recognizing that there are stochastic elements both in the determination of decisions and of decision outcomes" [Nelson and Winter, 1982].
Routines are the main method to take decisions and generate events in anti-neoclassical models, such as agent-based evolutionary economics models, embedding bounded rationality and uncertainty.
In short, routines are both feature of reality and of models, their interest laying in the potential export from one realm to the other.
The paper is organized as follows: we quickly remind some substantive cases where routines are used by consumers and firms. Then we enter into the field of computer routines for four chapters (their components, a list of types of routines with their typical code, the role of random factors inside, some example of routines used in our models).
After coping with some objections to the routine approch, we cover in three chapters the field of empirical evidence, proposing some reflections about the origin of routines that, totally obvious for the psychologists of early child development have been overlooked by economists, then broadly relying on existing literature on organizational routines, to finally sketch routines in interaction environments (across individuals and organizations).
After reviewing how to cope with dynamics (routine evolution, transmission and shift) we propose in four chapters a possible approach for a fruitful interconnection between artificial and real-world routines (proposing a "golden rule" for modelers emphasising empirical correspondence for computer routines to questionnaire items, a new method to explore the quality of the routine through a testbed in Excel format to inter-personally establish whether a certain artificial routine is successful in simulating actual data, a short remind of our inclusion in humans in computer models so as to elicit realistic routines and a research program for cumulative knowledge and empirical evidence on routines).
Accessible in language to the general public, the papers includes at least 4 innovative cutting-edge issues for agent-based experts.
Consumers use routines, called also "rules of behaviour" or "rules of thumb" to recognize their needs, to search for solutions and alternatives, to assess how well their needs would be satisfied by different combinations of products and services, to compare opportunities, to select acceptable choices and make final decisions, including repurchasing after experience.
Firms use decision making rules or routines to choose product design, their quality variants and to establish the production level, to maintain standards of production quality, to fix prices, to monitor competitors, to approach trade channels and final consumers, to advertise (or not!) their brands and products. They use production routines as "blueprint" for productions, to which a large number of tacit knowledge is attached, to make them operational.
Banks use routines e.g. to attract and select customers, provide loans, ask for anticipated repayments. Insurers use routines ("underwriting guidelines") to propose a customized premiums on request and decide whether to refuse to sell to some too-risky customer.
Voters use routines to select the source of information they rely upon, to attribute importance to electoral and political issues, to evalute competing parties' positions, to decide whom to vote and whether to vote at all.
In models, a routine is a computer code, e.g. in BASIC, PASCAL, LSD or other computer languages, whose components are fairly general and understandable without a specific knowledge of the language itself.
A routine is a set of operations, performed in a sequence, over variables, having inputs and outputs, which are connected through state values and parametres. They can be organized in one or more "stages" or phases. They include deterministic and stochastic (i.e. random) elements. Routines relate to agents (e.g. firms, people,...) and their interactions.
They have antecedents (other routines producing the variables and inputs used by the routine) and consequential routines (using their outcomes), while inside the routine there may be sub-routines.
For instance, a routine fixing a price offered by a firm might add a percentage to a standard cost. Antecedents are the computation of the standard cost and the percentage. Consequential routines could include the evaluation by potential customers who decide whether to buy or not.
Without any claim of exhaustiveness, we can single out the elements of routines in evolutionary economics ABS (agent-based simulation) models in the following table:
Random extractions from a variety of statistical distributionS, including the uniform distribution, Gaussian distribution and asymmetric distributions, can play several roles in evolutionary agent-based models:
1. they provide a source of heterogeneity and variety across the population, the parametres and the events occurring during the simulated time;
2. they cover the effect of missing variables, which have not been explicited in the model;
3. they allow the agents to have "private reasons" for their decisions, referring to irreducible singular connections and sources, which would be irrelevant to model, as they happens just once and for one agent, without the possibility of making "policies" and "strategies" about them;
4. they generate "historical" events during the simulation, allowing for path-dependency;
5. they produce novelties over time and allow future event not to be perfectly forecasted, which prevents agents from taking maximising decisions (e.g. to wait to buy until a certain price is at his lowest level), compelling to satisfycing decisions (e.g. to buy when the price is lower than a certain threshold);
6. they provide a powerful justification of the agents' ignorance of the deep structure of the model and their bounded rationality (if a model would consist only of deterministic and simple relations, an intelligent agent could finally guess how it works exactly);
7. they allow to avoid totally unlikely symmetric situations, too often assumed in neoclassical models for the sake of simplicity (e.g. consumers have all the same preferences and income, firms sharing exactly the same technology, etc.);
8. they insert a wedge between choice and actual fulfilment of goals and actions.
Consumers have been considered as having routines to decide whether to buy or not to buy, which one to buy (out of qualitatively differentiated version of a good), using three alternative routines, depening on a higher-order routine (which attribute one of them to a specific consumer largely depending on income).
Potential buyers of insurance have been given routines taking into account the risks they face, affordability of policy, and a sufficent probability of losses. Conversely, insurers have been given rules to refuse to sell to certain applicants.
Firms have been given routines to choose the price of the good they sell, the amount of Research and Development and its prevalent use, in order to obtain certain performance from product design, to decide whether to take a loan and when to reimburse debt (in the same model, humans can take the same decisions, so an external eye can try to elicit routines from their behaviour).
Exporters have been given rules to choose which foreign market to target and how to discriminate their prices in the different markets.
Routines are a key tenet of behavioural economics. They can be critised by two opposite fronts:
1. Some might say that people don't follow routines but rather make of "improvisation" the key feature of their decision-making process. They decide at the very last minute without too much consideration of many objective feature of the choice at hand but rather under the pressure of mood and others' opinion. Their choices seem not to be forecast and lack any ground of further justification. If in certain circumstances this can be an appropriate description of their behaviour, routines that would include multiple and large sources of randomness could capture this situation. At the same time, even with a very low level of rationality and coherence over time, certain aggregate properties could still emerge, as you can see in this model.
To an external eye, irrespective of private reasons and internal troubles in decision-making (doubts, memory lapses, incomplete knowledge,...), outcomes might be observed and referred to some inputs, history of the agent, and external conditions, with more or less stochastic factors to allow for wide uncertainty.
In other words, the word "routine" can be stretched down to a very lax requirement. In particular, it does not need to be excluding "innovation" and novelty.
2. On the opposite front, neoclassical theory claims that economic agents maximise objective functions (such as utility for consumers, profits for firms, votes for politicians, etc.) under constraints (e.g. budget, demand, voters' preferences,...) using fairly sophisticated mathematics (e.g. derivatives). Every experiment with real people denounces that this is not the case. People don't use such amount of calculus and mathematics nor have a recognizable function to maximise. There is overwhelming empirical evidence of bounded rationality, where people have cognitive and psychological constraints that prevent them to follow any high-level mathematics. The neoclassical theory is factually false, in those parts that are falsifiable.
Some defenders of the neoclassical theory claim that people behave "as if" they were rational maximisers of objective functions because if they weren't, somebody else would achieve more, showing what to do. For instance, firm must maximise profits because if they don't, other firms will be more profitable, investors would shift to them, and the former would have no choice but close or imitate the winners. Competition would do the trick.
But you can't behave as you were a genius of mathematics if you aren't. People don't have the skills and the inputs they would need to compute the maximisation and they cannot fake an ability they don't have. Try to sit an exam of mathematical topology without having any preparation: you'll not succeed!
And competition is among real agents: if everyone doesn't do something, competition will not work against any of them. You can't effectively choose options not on the table.
Certain routines may turn out to be too simplistic and, in principle, there may be a dynamics of change in routines towards higher sophistication and coverage of problems, but this is exactly one of the issues explored in the models and in the qualitative reflections of evolutionary economists.
Where the routines come from? People follow routines learned when they were children and perfectioned over the years. Newborns are subject to routines (such as cleaning) performed by adults to them. Later on, they learn to perform similar routines under supervision (e.g. lacing their shoes) and, further, interiorize them so as to repeat without supervision.
In parallel games, they sit with other kids with no interaction, or, later on, with imitation and conflict (e.g. for the same toys). In bilateral or team games, they play under a set of rules: little children want to win and often jump to conclusions without following the rules of the game, while older children take pride in winning through smart strategies and tactics, responsive or irresponsive to others' behaviours. They develop meta-rules for localizing the routines proper to the situation and the place (e.g. distinguishing what is to be done in bathrooms, dining rooms and open parks) as well as coping with breaching the rules, being punished and challenging the rule-setter.
In parallel to teaching and learning from others, children autonomously explore possibilities of movements and decisions, take risks and generate results. They may memorize situations and effects of their movements and activities, developing distinctive personality traits (e.g. curiosity or prudent risk-aversion) to broadly cover many activity settings (both in fictious games and in reality). Cognitive faculties are enhanced by story-telling, phantasies and dreams, emotionally charged and rich of moral insights.
Later on, as consumer, workers and voters, people will enact routines deeply entrenched in their earlier ages, developed over the years, crystalized in certain key moments and evolving in constant connection with external and internal environment.
Individuals (in their roles of consumers, workers, decision-makers but not only) use routines outspokenly, consciously or unconsciously:
Outspoken routines are declarations to others and commitments to decisions and choices. Conscious routines are non-spoken but intentional activities, often directed to a goal. Unconscious routines are short or long strips of activities activated automatically without efforts and along hardwired irreflective motions (e.g. muscolar movements).
In practical terms, to give an example, driving is a good intertwining of outspoken declarations ("Ok, now I have to go to work"), conscious routines (about the track and key focuses for starting and driving) and unconscious strands of many movements per minute, which are immediately forgotten.
In households, there is a lot of outspoken routines, where people exchange proposals verbally.
On a larger perspective, there is a broad correspondence of this trychotomization with the Freudian proposals of Super Ego, Ego and Es.
From a psychologist's point of view, routines are particularly important for an "obsessive" person, who rigidly follows "rituals", irrespective of their context and reasonability, often for escaping fears in "compulsive obsessive" types of behaviours. Then their polar opposite would be a "chaotic" type (which we would capture with an abudance of "random components").
The most difficult to model is a good balancing personality, expressing the self in an appropriate way in most contexts, flexibly alternating and adapting routines. A possible approach is to have a higher-level routine selecting which substantive routine has to be activated and an even-higher-level "context" routine to choose from two or more "selecting routine", as this picture describes:
For final outcomes, of relevance are not only the rules but also the parametres and the values of the current and past states, which can be present and and influence any of the routines.
"Large part of the tasks carried out in organizations, such as manufacturing, marketing and selling goods and services, are accomplished in routinized ways. This is not only true for trivial operations, such as manufacturing, but also pervades processes such as decision making, strategizing or even change and innovation. Organizational routines are the building blocks of organizations: they capture the typical ways in which organizations accomplish their tasks".
Organization use routine to make foreseeable decisions and reach consensus; routines can be mostly procedural on roles of decision-making (e.g. who is entitled to make the first proposal, who should be consulted, who take the final decision, who is implementing it, who gives feedbacks), timing (how often and in which period proposals are made, how much time is taken during the evaluation phase), management rules (what to do if somebody is delaying to perform its role, what do to when objections are raised without positive indications, etc.).
The roles routines seen to have in organisations include the following: to co-ordinate and control actions and decisions, to provide power 'truce' across the sub-units and sub-cultures within the firm (different departments, workers vs. management, people in multiple location of a multinational company,...), to economise on cognitive resources, reduce uncertainty, to lead to inertia, to provide stability and enable and constrain, act as triggers, and to embody knowledge [Becker, 2002]
Since the routines used within large or small organizations have been the object of a huge array of research, we shall be largely draw this chapter on literature, mainly Nelson and Winter, Becker 2008, 2009, 2002.
Routines are a persistent feature of the agent, which determines its possible behavior (though actual behavior is determined also by the environment); they are heritable over time and from an agent to another, and they are selectable in the sense that agents with certain routines may do better than others, and, if so, their relative importance in the population may be augmenting over time.
Routines play a key role in the development of an evolutionary explanation of economic change. Such an explanation works by identifying the mechanisms of variation, selection, and retention. Routines are supposed to be the unit of analysis. Therefore, one has to be able to explain how the variation of routines comes about, how routines are selected for, and how routines are retained over time.
Routines are 'pattern of behaviour that is followed repeatedly, but is subject to change if conditions change'
Organizational routines give legitimacy to decision-making, where the process and the outcome are strictly linked and influence each other.
A key tenet of evolutionary economics on routines is that their is a continuum between productive routines and decision-making routines: you can have dysfunctional routines in both domains and have innovations and improvements. Routines in production e.g. select which raw materials and parts should be combined, where and how to handle them, which pre-treatment and treatment they should receive, how to apply energy and chemicals to them (and whether there is such necessity), how to assembly and finish the product, how to package it, test it, develiver it to intermediaries and final clients.
In manufacturing, there is a constant effort to routinize and standardize such processes, so as to obtain large numbers of perfectly identical products to be sold, whereas in hand craftsmanship each product can be subject to a number of creative steps for a final uniqueness and finesse, with the application of largely tacit skills, embedded in humans, to certain (carefully evaluated) materials. There are many intermediate situations, with e.g. the so called "mass customerization" or repetitive craftsmanship. In other terms, sector-wide state-of-art production routines preveails until innovations are introduced and while resistant-to-change old habits persist.
The performance of an organizational routine involves the effective integration of a number of component subroutines (themselves further reducible), and is ordinarily accomplished without "conscious awareness" - that is, without requiring the attention of top management.
Routines can connect and rely on distributed knowledge held by different members of an organisation that does not completely overlap, and that it is very difficult, if not impossible, to get the overview over the 'whole' knowledge in the organisation. The distributed nature of routines gives rise to intransparency and complexity.
Like habits, routines are self-actuating, being executed in a virtually automatic manner. Reflection or volition is absent or not necessary. Like habits, routines are characterised by individuals following them without deliberation, without devoting conscious or explicit attention. We are not usually aware of them as long as they run smoothly, and only become aware of them when they do not.
An organization might be expected to encounter difficulty in departing from its prevailing routines, but it should have no trouble in conforming to them.
Several characteristics along which the processual nature of routines can be described have been identified in the literature: decay, leading to a need for 'maintenance' of routines; decay speed; the speed of executing routines, of changing their contents, and of switching between them; reaction speed; reaction time, time lags, and time delays; frequency of repetition and point of time of impact ; frequency and fashion of shifting from one routine or set of routines; age (duration) of an activity, speed of environmental change, quality of information with regard to the activity, amount of managerial and employee turnover, and volatility of the decision environment which all can act to intensify or dispel the influence of routines.
Organizational routines occupy 'the crucial nexus between structure and action, between the organisation as an object and organising as a process'.
Routines are embedded in an organisation and its structures and are specific to the context the application of general rules to specific contexts always involves incomplete specification and missing components, and thus the necessity of completing them. This will always require 'repair skills', such as interpretation and judgement skills, for example to know what routines to perform and when.
Because routines are embedded and interlinked, they are also able to identify supporting complementary elements that are necessary for their implementation in a specific context.
Several kinds of specificity have been identified in the literature: historical specificity, local specificity, and relation specificity. Historical specificity derives from the fact that whatever happens does so at a certain point of time, which is characterised by a certain constellation of environmental factors and interpretative mindsets. Because such constellations will be complex, there is a low probability that routines can be replicated identically. Specificities also arise because routines are the outcome of local learning processes. Local specificity also arises because of cultural differences and limits to generalisation arising from those.
The most important implication of specificity is that routines are transferable to other contexts to a limited extent only. When removed from their original context, routines may be largely meaningless, and their productivity may decline when transferred routines are path-dependent and shaped by history.
How they will develop is a function of where they have started out from. Based on their previous state, routines adapt to experience incrementally in response to feedback about outcomes'. Recognising the path-dependent nature of routines highlights the importance of feedback effects.
Organizational routine can be referred to three different concepts: recurrent behaviour patterns, rules or procedures, and dispositions (the assignment from a set of possible events to a subset of actual events). For more on this see Becker.
In short, routines seem to require some repetition, possibly connected to organizational incentives, limitation of knowledge and computational capability, and truce across conflicting units and power centres.
The final effect is a relatively stable and detectable pattern of decisions and material outcomes (products, etc.).
Routines express the way interaction across different agents occurs and generate outcomes. Certain locations (a given city, industrial district or countryside village) could be characterised by the over-proportional presence of certain routines in individuals, organizations and their connections.
For instance trust could be widely present across people working and living in the area, price signals could spread quickly, innovations could be thoroughly assessed and embraced ("industrial athmosphere" in industrial districts). Setting up new districts, by modifying the ease of these interactions, can be a goal of a modern industrial policy (by the way, this is one of my personal field of consulting activities, especially in the form of eco-neighbourhood and green industrial districts, where there is a concentration of supply and demand for eco-innovations, on the one hand, and of green jobs and green qualifications on the other).
If you enter into a Turkish bazaar, you'll enter an interaction space where goods and relations are exhibited, established and exchanged, finally leading, possibly and among other things, to a purchase. If you are from Turkey you'll be able to craft your own behaviour in the proper and effective manner, whereas if you are not used to bazaars, you'll probably behave according to unadequate routines and stereotypes.
Supermarkets are different interaction spaces, and Internet navigation as well.
Interaction spaces are at the intersect of different agents' boundaries and routines, the latter serving for regulating the connection, the communication, the transaction and the lasting effect of the conversation being possibly held. In this vein, routines open or close the conversation, allows for bargaining and delivery, etc.
In interaction spaces, routines are often "social norms", not established by any of the real agent present at the time of the interaction but embedded in a much larger and more stable framework, as well as in the interpretation the agent take of it.
Higher-order routines establish which specific routines to use (thus to which one to shift under certain circumstances), what influence their parametres, how they evolve, transmitted and are substituted by others.The dynamics of "creation", evolution, selection of routines can be inside the agent or can be linked to the enviroment, with "survival" of the fittest of the routines being one possibility, while too simple Pavlovian "learning" processes (based on repetion of external stimuli and rewards) may not capture the way humans develop their routines, as we underlined before.
Imitation of others' routines, the innovation of our own routines, the constructive criticism of others' routines to generate new ones, hinging on a positive ansure to the criticism, and the hybridization of routines coming from different sources (including our own) are the four main ways of routine evolution and change.
Changes in the proportion of routines over the total of agents depend on the introduction of new routines, their diffusion, the reduction and finally disappearance of (e.g. older) routines.
In many models, routines are fixed by the modelers, who is interested in seeing how certain outcomes are related to inputs, starting conditions and parametres. In other models, agents interally change certain parametres of their routines or, due to the change in values, select other routines to use. A third class of models hinges largely in studying the inter-agent generation, diffusion and selection of routines. In all models, it's very important to report not only preliminary results, but provide methods to systematically explore the robustness of statements and stylized facts.
In 2004, we proposed a "golden rule" constraining who produces a model with routines inside to select formal rules that can be converted into questions for real consumers in a questionnaire, such as those used in real-world market research.
Our rules should be convertible into simple questions that a normal consumer can answer.
This "golden rule" has two advantages:
After publication, this proposal has been evaluated as one of the main future venues for research and implementation into agent-based models in a peer-reviewed article published by Technovation journal. Brent Zenobia et at. write: "Piana (2004) argues that it should always be possible to convert the behavioral rules of consumer agents into questions for real consumers in questionnaires. This is important to achieve empirical feedback, calibration, and validation of the Agent Model, as well as to help ensure that the agent specifications do not become so contrived that they defy description in ordinary terms. At present, few methodologies exist to help link consumer questionnaires to agent behavioral rules; more research is needed in this area".
In order to elicit new routines actually used by real people (or at least being acceptable in simulating actual decisions), we have developed over the years several agent-base models in which humans take a part and play (as they were a sort of "videogame").
Give a look to this model of monopolist (2001) and this dynamic competition between two humans or a human agains an artificial agent (2003), a similar approach of "partecipatory simulation" in Hanckock (2010), a review of several "Participatory agent-based modeling" in Li An (2011) or the 2006 statement by Duffy that "data from human subject experiments provide a ready-made source of empirical regularities that can be used to calibrate or test ACE models of individual decision-making and belief or expectation formation". As you shall see from the publication years, we were among the first to use humans in agent-based models.
In so doing, we and our colleagues directly violated the definition of the The New Palgrave Dictionary of Economics (Second Edition, 2008) that states that "Agent-based models consist not of real people but of computational objects that interact according to rules".
Our purpose is to allow humans to take decisions, develop strategies and tactics, experiment and learn, offering "outspoken routines" and conscious choices to be offered to the attention of the modeller.
By feeding back those rules into artificial agents one could obtain better and more realistic models.
In this paper we present a method of inter-personal validation of routines, as capable of generating data that are structurally similar to real ones.
More specifically described in this Excel file available for download, the method consists in asking others to recognize a column of data as coming from the artificial routine ("A") or from the real world ("R") and evaluting their answers' correctness and timing.
Artificial routines are considered as realistic if there are many mistakes in attribution and if the time to answer is long. In parallel, the opinion of the participant is asked about the difficulty of the task. Inter-personal validation is achieved by asking participants (ignoring the overall method of evaluation and separated by the experimenter, who introduced the routine into the program).
People can recognize patterns, without being anticipated on what to look for. If at their eyes there is no structural difference between columns from the real world and from the artificial routine, then the routine is good in simulating reality, the more so since we in economics look at the behaviour of people like them.
Sophisticated statistical tests of "goodness of fit" are not used by humans in their choices - if they don't see difference in front of the Excel column, they will not notice the difference in front of a supermaket shelf.
This procedure of pattern recognition can be traced back to many empirical experiments in the psychology field, and can be tracked back to the famous Turing Test (1950) for artificial intelligence, but in economics - and for Agent-based models - our proposal is new and, we hope, useful for researchers.
In the Excel file, we introduces several columns of real data about prices in a supermarket, while generating other columns with a routine activated by a button. Experiment with it and make your reflections on the method, while asking others to recognize columns.
You can introduce your own real data and your own routines, by changing the very simple code in the buttons. Please let us know of your experiments and thoughts!
Building on what we described in our dissertation at Bocconi University in 1994, we underline that the comparison between models and the real world can be done at three levels:
1. at the level of the single routine;
2. at the level of single agent, checking whether all key decisions usually faced are covered by routines;
3. at the level of the interaction between (groups of) agents, checking whether all key agents have been coveredand whether there are relationships and flows of information missing.
On the first level, we must check whether the single routine is realistic or simplistic. Conversely, it's crucial to have an Internet freely accessible site where an open list of decision-making routines are collected from empirical evidence. Through interviews, questionanairs and direct or participating observation we should highlight a good number of approaches and routines, based on different set of information and arguments. At this stage, the relative frequence of use is not important: we want the longest possible list of routines used in different parts of the real world, including sectors, countries, organization size, etc. Indeed, it's fully possible that rules are different among mono-product and multi-product firms, entrants and incumbents, marginal or core entreprises, financially well-established or fragile, technological cutting-edge or laggards, etc. In the list, what's important is the detection anywhere and anyhow.
The second step is to organize this variety, trying to group similar rules, to identify the reasons that supersede over the choice of rule instead of another, to establish whether the rule tend to be applied sistematically or erratically, for long or short periods of time, etc. When does a rule be changed? Is any imitation taking place? In this second stage, you have the problem of the co-existence on the same market of agents using different rules and their relative frequence. Determinants of the frequency distribution, the reasons for changes in the frequency distribution, the potential for evolution and variety and what can influence the values of the parametres within the routines are all typical issues of interest.
Such cumulation of accessible knowledge of routines, possibly expressed in computer code and rationalised, searchable according their inputs and outputs and other key elements would be extremely important to develop realistic agent-based models.
Our testbed could be used to quantitative judge the different routines, especially if sets of real-world data could be openly accessed.
The first comparison would be between the routine used in the model and the list in the database. If it's not there, we shall know that there is a problem, to be solved using what Thomas Kuhn calls the "normal science". But it's fully possible that routine is not only there but is a good representativ of an entire class of routines using the same inputs or structure.
As a second step, you compare the empirical results with the hypothesis of the model concerning the stability of the routine over time and across agents, the degree of approximation with which it is followed, and the frequency of the rule across different firms and sectors.
You should try to understand whether the real agents use more sophisticated or more simple rules, if their knowledge base actually includes the required inputs, etc. It's fully possible that the rule is adequate for certain countries and sectors, while being at odds with others. This is key for "localizing" the model, which applies where the routines it implements are actually used. The model is valid for the part of reality where there is a structural similarity between the routines of the model and the routines used by real agents.