Sie sind auf Seite 1von 4

Economics Letters 175 (2019) 15–18

Contents lists available at ScienceDirect

Economics Letters
journal homepage: www.elsevier.com/locate/ecolet

Prior knowledge and monotone decision problems



Bernard Sinclair-Desgagné
HEC Montréal, Canada

highlights

• Prior capability to behave optimally when observing an expected signal gives rise to automatic responses or routines, whereby elementary actions are
tied together and triggered through a graded control.
• A rational decision-making entity with such a priori skills will resort to a routine if this allows it to observe a relevant signal.
• This transforms the initial decision problem into a Monotone Decision Problem.
• Prior know-how can thus have a key influence on the way economic actors formulate and make their choices.

article info a b s t r a c t

Article history: This paper identifies cognitive conditions under which a rational individual or group decision-maker will
Received 6 September 2018 want to commit ex ante to some choice restrictions, in order to get extra information about an uncertain
Received in revised form 27 November 2018 state of nature. We show that the implemented limitations will then bring her to solve a monotone
Accepted 28 November 2018
decision problem. This provides a behavioral/microeconomic foundation for the recurrence of this type
Available online 4 December 2018
of problem in many contexts. From another angle, the analysis explains why and when individuals and
JEL classification: organizations resort to automatic responses or routines in some circumstances, and how this can shape
D01 their environment.
D21 © 2018 Elsevier B.V. All rights reserved.
D80
D83
Keywords:
Choice architecture
Attention and effort
Limited attention
Design attributes
Routines

1. Introduction formats a subject to act upon evidence in a certain way. The issue
matters as well for organizations: firms must somewhat tie to-
When a decision’s payoff is subject to uncertainty, to what ex- gether their business units (especially those respectively in charge
tent would a rational individual or group decision-maker accept to of product design and marketing) if they want to effectively gather
limit her choices against the possibility of obtaining extra informa- and process relevant business intelligence (Gold et al., 2001).
tion? At first sight, the answer rests on the value this information Oftentimes, the nature of the obtainable signals – be they
brings; but this in turn depends on what the constrained decision- numerical, like prices, or subject to measurement – will allow
maker will be capable of doing once she gets the additional data.1 a decision-maker to rank them. Suppose, then, that the decision-
Much of individual or team training involves this precise tradeoff: maker knows a priori (or believes she knows) how to best react
training enables someone to get hold of relevant info, but it also to an observed signal. This characterizes a situation with so-called
design attributes, where “(...) there is a great deal of a priori informa-
∗ Correspondence to: Environmental Economics and Global Governance Chair, tion about the form of the optimal solution, that is, about how the
Department of International Business, HEC Montréal, 3000 Chemin de la Côte- variables should be related.” (Milgrom and Roberts, 1992, p. 90) In
Sainte-Catherine, Montréal, Canada H3T 2A7 . those circumstances, this paper shows that, in order to get relevant
E-mail address: bsd@hec.ca. information, the decision-maker will consent on having her choice
1 For this reason, the issue considered here does not fit the framework of real
set configured as a list of graded moves.
options (for some influential presentations on this subject, see Dixit and Pindyck,
1994, and Trigeorgis, 1996). In the latter, a decision-maker weighs up the possibility
This result highlights two (objective or subjective) contextual
of making a commitment after receiving information; here, she must commit before features – rankable signals and design attributes – that will
information comes in. lead a rational decision-maker to precommit to view and handle

https://doi.org/10.1016/j.econlet.2018.11.030
0165-1765/© 2018 Elsevier B.V. All rights reserved.
16 B. Sinclair-Desgagné / Economics Letters 175 (2019) 15–18

an upcoming situation as a Monotone Decision Problem (MDP). the probability space (Ω , F , P), where F is a σ -algebra of subsets
Recall that a decision problem is monotone if the decision-maker’s (events) of Ω , and P : F → [0, 1] is a probability measure. Assume
observation of a higher signal leads her to choose a higher ac- that, for every x, the composite application u ◦ Y (x; ·): Ω → Υ
tion. Problems of this sort have been recognized in many settings, is integrable with respect to P; the decision-maker’s choice then
such as production planning, financial management, insurance, evolves around the expected valuation EP [u (Y (x; ω))], where EP
auctions, contracting, and organizational design. Economists thus refers to the expectation operator based on P.
keep devoting substantial research effort to their analysis.2 On a For some measurable space (H , Γ ), where H is an arbitrary set
different note, this paper offers a first explanation for their preva- and Γ is a σ -algebra of subsets of H, let Θ : Ω → H be a
lence. measurable mapping taking values in H. We interpret a realization
One can also see this paper’s result as a statement address- θ = Θ (ω) as a signal about the state of the world ω. With no loss of
ing the interaction between automatic (or rule-based/‘system 1’) generality, we assume that Θ is surjective (or onto), i.e. Θ (Ω ) = H.
and controlled (or adaptive/‘system 2’) decision processes. Dating The decision-maker could readily seek an element of X that
back at least to Simon (1947)’s and Cyert and March (1963)’s maximizes EP [u (Y (x; ω))]. Alternatively, she could first commit
respective accounts of the cognitive limitations of individuals and △
organizations, there is a sizable and growing literature dealing to a restricted choice set X ⊂ X in exchange for the ability to

with bounded rationality and limited attention (see, e.g., Conlisk, observe Θ (ω), then choose in X a maximizer of the conditional (or
1996; Rubinstein, 1998; Dessein et al., 2016, and the references revised) expected valuation EP [u (Y (x; ω)) | Θ (ω)]. Which path
therein). This literature now draws largely on psychology and would she go for? Two key notions must be defined before we
the neurosciences (e.g., Kahneman, 1973, 2003, 2011; Angner and provide a formal answer to this question.
Loewenstein, 2007; Camerer et al., 2005; Loewenstein et al., 2008;
Recall that a chain is a totally ordered set.
Fehr and Rangel, 2011). It notably considers the fact that individ-
uals and organizations often sacrifice on extensive deliberation
Definition 1. A routine is an injective (or one-to-one) function
in exchange for grasping their environment better. One type of → →
behavior – which is the one mostly studied in current research r : S → X , where the domain S of r is a chain.
– consists in economizing on information acquisition and pro-
This definition renders the idea conveyed in the introduction
cessing: decision-makers first settle on an information structure,
that a routine is like a dial, a knob, a switch or a tuning device. The
then play the ensuing ‘cognitive game’ (Tirole, 2014). This leads to → →
using multinonial logit models in discrete choices contexts with chain S stands for a graduated ring or strip. Once a spot s on S is
imperfect information (Matĕjka and McKay, 2015), employing the hit, specific values of the decision variables are triggered through

monotone likelihood ratio property to order noisy signals in auc- r(s), and no other spot s′ on S activates exactly the same values.
tions and global games (Mensch, 2018), and transforming Bayesian The second notion now evokes some way to represent the
games with strategic complementarities into two-action games objects of a set.
(Amir and Lazzati, 2016). Another type of behavior – which this pa-
per looks at – would rather develop and rely on adequate reflexes, Definition 2. A set A parametrizes a set B if there is a bijective (or
production habits, standard operating procedures, or routines, the one-to-one and onto) function p : A → B.
latter meaning that “(...) the parts of an individual’s skill which are
completely routinized are the parts that he or she does not have A parameter a ∈ A thus acts as a label, a code name, or a sticker
to think about — once a routine is switched on in the worker’s uniquely ascribed to an object p(a) = b ∈ B. Knowing a, one knows
mind, it goes on to end without further consultation of the higher b, and vice-versa.
faculties.” (Nelson and Winter, 1982, p. 63) We model this conduct
as having the choice set itself changed into a list of labeled subsets 3. Main result
of specific values that basic decision variables can take; once a
given label is picked, all these variables are determined. According
The issue raised at the start of this paper can now be answered
to our main proposition, a rational decision-maker with enough
as follows.
prior ability would stick to such a scheme in order to grab an
informative signal.3 →
The upcoming section lays down the mathematical background Theorem. Suppose that (i) a chain H parametrizes the set of signals
for this result. Section 3 contains the formal proposition and its H, and (ii) for any θ ∈ H, arg maxx∈X EP [u (Y (x; ω)) | Θ (ω) =
demonstration. A concluding Section 4 conveys further remarks θ] ̸= ∅. Then, the decision-maker should find it preferable to rely on
and observations. a routine if she can thereby gather Θ (ω).

2. Notation and definitions Proof. Observe that, for all x ∈ X ,

EP [u (Y (x; ω))] = EP [EP [u (Y (x; ω)) | Θ (ω)]]


Consider a decision-making entity (which can be an individual [ ]
or an organization) with preferences over a set of outcomes y ∈ Υ ≤ EP max EP [u (Y (x; ω)) | Θ (ω)] .
x∈ X
represented by a valuation (utility, profit, welfare) function u : Υ
→ R. The outcomes stem from a transformation Y : X × Ω → Υ , Now, consider the subset 𭟋 of H × X defined as 𭟋 = {(θ , x) |
an instance of it is written y = Y (x; ω), where x ∈ X denotes the x ∈ arg maxx∈X EP [u (Y (x; ω)) | Θ (ω) = θ] for some θ ∈ H }.
decision variables and ω ∈ Ω is an uncertain state of the world. By the Axiom of Choice,4 there exists a ‘uniformization’ of 𭟋, i.e. a
The decision-maker’s prior beliefs about the latter are encoded in ‘uniformizing function’ f : H → X such that, for all θ ∈ H, the

2 Among many others, especially in the literature on economic complementari- 4 The Axiom of Choice is a key (yet debated) axiom of Set Theory. Its most usual
ties, see Athey and Levin (2018) and the references therein. statement runs like this: Given a non-empty family à = {Ai }i∈I of non-empty sets,
3 In this paper, the words ability, skills, capabilities or know-how – rather there exists a ‘choice function’ for Ã, i.e. a map f : I → ∪i∈I Ai such that f (i) ∈ Ai
than information, understanding or erudition – are the appropriate synonyms for for all i ∈ I. In the present case, we take H as I and each set 𭟋θ = {(θ, x) | x ∈
‘knowledge’. arg maxx∈X EP [u (Y (x; ω)) | Θ (ω) = θ ]} as Ai .
B. Sinclair-Desgagné / Economics Letters 175 (2019) 15–18 17

→ →
pair (θ , f (θ )) ∈ 𭟋.5 Let r̄ = f ◦ p: H → X , where p: H → H is the cultivate the habit of thinking of what we are doing. The precise
→ opposite is the case. Civilization advances by extending the
assumed bijection between H and H. Clearly,
number of operations which we can perform without thinking
EP [max EP [u (Y (x; ω)) | Θ (ω)]] about them.
x∈X
= EP [max EP [u (Y (r̄ (s) ; ω)) | Θ (ω)]] .

s∈ H
4. Concluding remarks

The function r̄ takes values in X and its domain is a chain. But


When a decision’s payoff is subject to uncertainty, we showed
it may not be one-to-one. We can nevertheless use it to construct
that a rational decision-making entity (an individual or an or-
a routine as the notion→is defined above.
ganization) could set the matter so as to be facing a monotone
Take the subset of H × X defined as {(s, x) | x = r̄(s)}. Invoking
decision problem. Sufficient conditions for this to occur are that
again the Axiom of Choice, there exists a uniformization of this
→ the entity be ex ante capable of (i) holding prior expectations on the
subset, i.e. a partial function r −1 : X → H with domain the range additional information she could get, (ii) encoding and ranking the
→ → →
of r̄ and image a subset S of H. Its right inverse r: S → X , a observed extra signals (by having them expressed as real numbers,
function such that r −1 ◦ r = id, is injective by construction, and for instance), and (iii) using any received signal optimally (thanks

S is a chain. This makes r a routine in the sense of Definition 1. to design attributes). In exchange for the possibility of getting an
Since r̄(s) = r(s), we have that informative signal, the entity will then commit to use a modified
choice set made of graded items pointing at specific values of the
max EP [u (Y (r̄ (s) ; ω)) | Θ (ω)] basic decision variables; she will thereby formulate an MDP.

s∈ H This result constitutes a first explanation for the repeated oc-
= max EP [u (Y (r (s) ; ω)) | Θ (ω)] . currence of monotone decision problems in economics. It suggests

s∈ S that such problems would necessarily (not exclusively, of course)
Taking stock of this relationship, the above inequality becomes arise in stable recurrent circumstances, when a priori capabilities
[ ] or know-how (via design attributes) are significant. They stem
EP [u (Y (x; ω))] ≤ EP max EP [u (Y (r (s) ; ω)) | Θ (ω)] . from the desire to then take advantage of this cognitive state, by

s∈ S
implementing reliable routines that economize on deliberation so
as to enhance external awareness.
The latter holds for any x ∈ X , so
[ ] This ‘attention strategy’, so to speak, fits the rapidly grow-
ing economic literatures on rational inattention (see, e.g., Gabaix,
max EP [u (Y (x; ω))] ≤ EP max EP [u(Y (r (s) ; ω)) | Θ (ω)] , 2014; Steiner and Stewart, 2016; Caplin et al., 2018; Ellis, 2018)
x∈X →
s∈ S and choice overload (see, e.g., Masatlioglu et al., 2012; Ortoleva,
which confirms the theorem’s assertion. ■ 2013; Manzini and Mariotti, 2014). It adds to these literatures in at
least three ways. First, while the latter focus on how agents deal
Under proper conditions, a rational decision-maker will then with information overload, it rather emphasizes the fundamen-

commit to focus her choices on the function r(·)’s domain S , tal tradeoff between internal deliberation and external alertness.
provided she expects to thereby get a valuable signal. This turns Second, while the latter come up with strategies based on coarser
her initial decision problem into facing a number of contingent information partitions or smaller choice sets, it rather stresses the
problems of the form presence of automatic processes and routines as opposed to con-
trolled actions (in line with Kahneman, 2011’s recent distinction
max EP [u(Y (r (s) ; ω)) | Θ (ω)] . (MDP)
→ between ‘system 1’ and ‘system 2’ modes of thinking). Third, while
s∈ S
the latter mostly consider a rational consumer, it rather applies
Since, by assumption, all possible signals θ = Θ (ω) lay on the to a rational producer, highlighting abstraction and parsimony on

graded scale H, and what is now the relevant set of decision input definition (via linking together and ‘black boxing’ the most

variables S is aligned on this scale (as the proof shows), problem elementary or trivial tasks) as a frequent mean to economize on
(MDP) is indeed a Monotone Decision Problem. attention.
Behaviorally speaking, solving (MDP) amounts to selecting the The construction method invoked in the Theorem’s proof was
appropriate level of a control device which automatically adjusts deliberately abstract; the intent is to thereby recoup as many
a collection of primary inputs or tasks. (Taking an example from situations as possible. Our framework still suggests some interest-
manufacturing, think of a just-in-time scheme which tightly co- ing applications, though. For instance, we were rather unspecific
ordinates procurement, production and logistics-related tasks ac- about the implemented routine, apart from the requirement that it
cording to specific demand forecasts.) The Theorem shows that this should point at contingent value-maximizers. If several such max-
somewhat mechanical modus operandi would be preferred by the imizers exist, then two decision-making entities holding similar
decision-making entity, as it enables her to learn about the state of preferences and beliefs could well adopt different routines, hence
nature. This agrees with a provocative assertion made some time exhibit a variety of working habits. This might explain the diversity
ago by Alfred North Whitehead (1911): of ‘styles’ which can be observed in people playing sports, driving
cars, giving talks, attending consumers, etc. Convergence on style
(...) it is a profoundly erroneous truism, repeated by all copy- among an organization’s workers is then often seen as a key feature
books and by eminent people making speeches, that we should of corporate culture (Kreps, 1990). Additionally, were the first
inequality in the above proof to be strict, there would be room
5 Uniformization is a major topic in Descriptive Set Theory (DST), a branch of for suboptimal routines. This might support an amount of conser-
mathematical logic which studies sets that are ‘well-behaved’ (so they avoid, in vatism or inertia in maintaining obsolete production habits up to
particular, certain weird consequences of the Axiom of Choice). Such sets are usually
a certain point, hence account for the presence of X-inefficiency
complete separable metric spaces (like the set R of real numbers) called Polish
spaces. Good introductions to DST can be found in Moschovakis (2009), Sion (1960), (Leibenstein, 1966) and low-hanging fruits (Gabel and Sinclair-
and Zapletal (2005). Desgagné, 1998) in organizations.
18 B. Sinclair-Desgagné / Economics Letters 175 (2019) 15–18

Acknowledgments Gabel, H.L., Sinclair-Desgagné, B., 1998. The firm, its routines, and the environment.
In: Folmer, H., Tietenberg, T. (Eds.), The International Yearbook of Environmen-
tal and Resource Economics 1998/1999 - A Survey of Current Issues. Edward
This note builds on my earlier work with Landis Gabel and An-
Elgar.
toine Soubeyran. It also owes substantially to several conversations Gold, A.H., Malhotra, A., Segars, A.H., 2001. Knowledge management: an organiza-
I had over the years with some friends and colleagues, including tional capabilities perspective. J. Manage. Inf. Syst. 18 (1), 185–214.
Bentley Macleod, Pascale Viala, Lars-Hendrik Röller, Jean-Pierre Kahneman, D., 1973. Attention and Effort. Prentice-Hall.
Ponssard, Pierre Lasserre, Marcel Boyer, Tracy Lewis, John Conlon, Kahneman, D., 2003. Maps of bounded rationality: psychology for behavioral eco-
nomics. Amer. Econ. Rev. 93 (5), 1449–1475.
Patrick Cohendet, Éric Giraud-Héraud, Pauline Barrieu, Louis Lévy- Kahneman, D., 2011. Thinking, Fast and Slow. Penguin Books.
Garboua, Claude Montmarquette, Thomas Boraud, Emmanuel Pe- Kreps, D., 1990. Corporate culture and economic theory. In: Alt, J., Shepsle, K. (Eds.),
tit, and Murat Yildizoglu. Valuable suggestions from an anony- Perspectives on Positive Political Economy. Cambridge University Press.
mous referee and the editor, Joseph Harrington, helped improve Leibenstein, H., 1966. Allocative efficiency vs. X-efficiency. Amer. Econ. Rev. 56 (3),
392–415.
the presentation. I am, of course, sole responsible for all remain-
Loewenstein, G., Rick, S., Cohen, J.D., 2008. Neuroeconomics. Annu. Rev. Psychol. 59,
ing deficiencies. I wish to thank the GREThA of the University 647–672.
of Bordeaux for its hospitality (under the IDEX-Bordeaux Visiting Manzini, P., Mariotti, M., 2014. Stochastic choice and consideration sets. Economet-
Scholars Program) while this piece was being written. rica 82 (3), 1153–1176.
Masatlioglu, Y., Nakajima, D., Ozbay, E.Y., 2012. Revealed attention. Amer. Econ. Rev.
102 (5), 2182–2205.
References Matĕjka, F., McKay, A., 2015. Rational inattention to discrete choices: a new foun-
dation for the multinomial logit model. Amer. Econ. Rev. 105 (1), 272–298.
Amir, R., Lazzati, N., 2016. Endogenous information acquisition in Bayesian games Mensch, J., 2018. Rational inattention and the Monotone Likelihood Ratio Property,
with strategic complementarities. J. Econom. Theory 163, 684–698. Working Paper.
Angner, E., Loewenstein, G., 2007. Behavioral economics. In: Gabbay, D., Thagard, P., Milgrom, P.R., Roberts, J.D., 1992. Economics, Organization and Management.
Woods, J. (Eds.), Philosophy of Economics, Vol. 13. Elsevier. Prentice-Hall.
Athey, S., Levin, J., 2018. The value of information in monotone decision problems. Moschovakis, Y.N., 2009. Descriptive Set Theory. Lecture notes available online.
Res. Econ. 72, 101–116. Nelson, R., Winter, S., 1982. An Evolutionary Theory of Economic Change. Belknap
Camerer, C., Loewenstein, G., Prelec, D., 2005. Neuroeconomics: How neuroscience Press.
can inform economics. J. Econ. Lit. 43, 9–64. Ortoleva, P., 2013. The price of flexibility: towards a theory of thinking aversion. J.
Caplin, A., Dean, M., Leahy, J., 2018. Rationally inattentive behavior: characterizing Econom. Theory 148, 903–934.
and generalizing Shannon entropy, National Nureau of Economic Research Rubinstein, A., 1998. Modeling Bounded Rationality. MIT Press.
(NBER) Working Paper No. 23652. Simon, H.A., 1947. Administrative Behavior: A Study of Decision-Making Processes
Conlisk, J., 1996. Why bounded rationality? J. Econ. Lit. 34, 669–700. in Administrative Organization, first ed. Macmillan.
Cyert, R.M., March, J.G., 1963. A Behavioral Theory of the Firm, first ed. Prentice-Hall. Sion, M., 1960. On the uniformization of sets in topolocical spaces. Trans. Amer.
Dessein, W., Galeotti, A., Santos, T., 2016. Rational inattention and organizational Math. Soc. 96 (2), 237–245.
focus. Amer. Econ. Rev. 106 (6), 1522–1536. Steiner, J., Stewart, C., 2016. Perceiving prospects properly. Amer. Econ. Rev. 106 (7),
Dixit, A., Pindyck, R., 1994. Investment under Uncertainty. Princeton University 1601–1631.
Press. Tirole, J., 2014. Cognitive games and cognitive traps, Working paper.
Ellis, A., 2018. Foundations for optimal inattention. J. Econom. Theory 173, 56–94. Trigeorgis, L., 1996. Real Options: Managerial Flexibility and Strategy in Resource
Fehr, E., Rangel, A., 2011. Neuroeconomic foundations of economic choice - recent Allocation. MIT Press.
advances. J. Econ. Perspect. 25 (4), 3–30. Whitehead, A.N., 1911. An Introduction to Mathematics. Home University Library.
Gabaix, X., 2014. A sparsity-based model of bounded rationality. Q. J. Econ. 129 (4), Zapletal, J., 2005. Modern Descriptive Set Theory. Lecture notes available online.
1661–1710.

Das könnte Ihnen auch gefallen