Professional Documents
Culture Documents
The Econometrics of Networks 1St Edition Aureo de Paula Online Ebook Texxtbook Full Chapter PDF
The Econometrics of Networks 1St Edition Aureo de Paula Online Ebook Texxtbook Full Chapter PDF
The Econometrics of Networks 1St Edition Aureo de Paula Online Ebook Texxtbook Full Chapter PDF
https://ebookmeta.com/product/chess-explained-
the-c3-sicilian-1st-edition-sam-collins/
https://ebookmeta.com/product/starting-out-the-c3-sicilian-1st-
edition-john-emms/
https://ebookmeta.com/product/the-paper-issue-83-1st-edition-
origamiusa/
https://ebookmeta.com/product/the-american-
revolution-1774-83-2nd-edition-daniel-marston/
Essentials of Econometrics 5th Edition Damodar Gujarati
https://ebookmeta.com/product/essentials-of-econometrics-5th-
edition-damodar-gujarati/
https://ebookmeta.com/product/the-econometrics-of-multi-
dimensional-panels-2nd-edition-laszlo-matyas/
https://ebookmeta.com/product/cancer-metabolomics-2018-paula-
guedes-de-pinho-editor-marcia-carvalho-editor-joana-pinto-editor/
https://ebookmeta.com/product/the-passionate-heart-1st-edition-
paula-quinn/
https://ebookmeta.com/product/physical-chemistry-thermodynamics-
structure-and-change-peter-atkins-julio-de-paula-tenth-2014-10th-
edition-peter-atkins/
THE ECONOMETRICS OF
NETWORKS
ADVANCES IN ECONOMETRICS
Series editors: Thomas B. Fomby, R. Carter Hill, Ivan
Jeliazkov, Juan Carlos Escanciano, Eric Hillebrand,
Daniel L. Millimet, Rodney Strachan
Previous Volumes
Volume 21 Modelling and Evaluating Treatment Effects in Econometrics – Edited by
Daniel L. Millimet, Jeffrey A.Smith and Edward Vytlacil
Volume 22 Econometrics and Risk Management – Edited by Jean-Pierre Fouque,
Thomas B. Fomby and Knut Solna
Volume 23 Bayesian Econometrics – Edited by Siddhartha Chib, Gary Koop, Bill
Griffiths and Dek Terrell
Volume 24 Measurement Error: Consequences, Applications and Solutions – Edited by
Jane Binner, David Edgerton and Thomas Elger
Volume 25 Nonparametric Econometric Methods – Edited by Qi Li and Jeffrey S.
Racine
Volume 26 Maximum Simulated Likelihood Methods and Applications – Edited by R.
Carter Hill and William Greene
Volume 27A Missing Data Methods: Cross-sectional Methods and Applications – Edited
by David M. Drukker
Volume 27B Missing Data Methods: Time-series Methods and Applications – Edited by
David M. Drukker
Volume 28 DSGE Models in Macroeconomics: Estimation, Evaluation and New
Developments – Edited by Nathan Balke, Fabio Canova, Fabio Milani and
Mark Wynne
Volume 29 Essays in Honor of Jerry Hausman – Edited by Badi H. Baltagi, Whitney
Newey, Hal White and R. Carter Hill
Volume 30 30th Anniversary Edition – Edited by Dek Terrell and Daniel Millmet
Volume 31 Structural Econometric Models – Edited by Eugene Choo and Matthew
Shum
Volume 32 VAR Models in Macroeconomics — New Developments and Applications:
Essays in Honor of Christopher A. Sims – Edited by Thomas B. Fomby,
Lutz Kilian and Anthony Murphy
Volume 33 Essays in Honor of Peter C. B. Phillips – Edited by Thomas B. Fomby,
Yoosoon Chang and Joon Y. Park
Volume 34 Bayesian Model Comparison – Edited by Ivan Jeliazkov and Dale J. Poirier
Volume 35 Dynamic Factor Models – Edited by Eric Hillebrand and Siem Jan Koopman
Volume 36 Essays in Honor of Aman Ullah – Edited by Gloria Gonzalez-Rivera, R.
Carter Hill and Tae-Hwy Lee
Volume 37 Spatial Econometrics – Edited by Badi H. Baltagi, James P. LeSage and R.
Kelley Pace
Volume 38 Regression Discontinuity Designs: Theory and Applications – Edited by
Matias D. Cattaneo and Juan Carlos Escanciano
Volume 39 The Econometrics of Complex Survey Data: Theory and Applications –
Edited by Kim P. Huynh, David T. Jacho-Chávez and Guatam Tripathi
Volume 40A Topics in Identification, Limited Dependent Variables, Partial Observability,
Experimentation, and Flexible Modelling Part A – Edited by Ivan Jeliazkov
and Justin L. Tobias
Volume 40B Topics in Identification, Limited Dependent Variables, Partial Observability,
Experimentation, and Flexible Modelling Part B – Edited by Ivan Jeliazkov
and Justin L. Tobias
Volume 41 Essays in Honor of Cheng Hsiao – Edited by Tong Li, M. Hashem Pesaran
and Dek Terrell
ADVANCES IN ECONOMETRICS VOLUME 42
THE ECONOMETRICS OF
NETWORKS
EDITED BY
ÁUREO DE PAULA
University College London, UK
ELIE TAMER
Harvard University, USA
MARCEL-CRISTIAN VOIA
University of Orléans, France
United Kingdom – North America – Japan
India – Malaysia – China
Emerald Publishing Limited
Howard House, Wagon Lane, Bingley BD16 1WA, UK
No part of this book may be reproduced, stored in a retrieval system, transmitted in any
form or by any means electronic, mechanical, photocopying, recording or otherwise without
either the prior written permission of the publisher or a licence permitting restricted copying
issued in the UK by The Copyright Licensing Agency and in the USA by The Copyright
Clearance Center. Any opinions expressed in the chapters are those of the authors. Whilst
Emerald makes every effort to ensure the quality and accuracy of its content, Emerald
makes no representation implied or otherwise, as to the chapters’ suitability and application
and disclaims any warranties, express or implied, to their use.
SECTION 1
IDENTIFICATION OF NETWORK MODELS
Chapter 1 Identification and Estimation of
Network Models with Heterogeneous
Interactions
Tiziano Arduini, Eleonora Patacchini and Edoardo
Rainone
SECTION 2
NETWORK FORMATION
Chapter 4 Trade Networks and the Strength
of Strong Ties
Áureo de Paula
SECTION 3
NETWORKS AND SPATIAL ECONOMETRICS
Chapter 6 Implementing Faustmann–
Marshall–Pressler at Scale: Stochastic
Dynamic Programming in Space
Harry J. Paarsch and John Rust
SECTION 4
APPLICATIONS OF FINANCIAL NETWORKS
Chapter 10 Financial Contagion in Cross-
holdings Networks: The Case of Ecuador
Pablo Estrada and Leonardo Sánchez-Aragón
Index
INTRODUCTION
ECONOMETRICS OF NETWORKS
Áureo de Paula, Elie Tamer and Marcel Voia
REFERENCES
Bramoulle, Y., Djebbari, H., & Fortin, B. (2009). Identification of peer effects through social
networks. Journal of Econometrics, 150, 41–55.
Calvo-Armengol, A., Patacchini, E., & Zenou, Y. (2009). Peer effects and social networks in
education. The Review of Economic Studies, 76(4), 1239–1267.
Chan, T. J. (2020). Snowball sampling and sample selection in a social network. In
Advances in Econometrics (Vol. 42).
Chen, H., & Strathearn, M. (2020). A spatial panel model of bank branches in Canada. In
Advances in Econometrics (Vol. 42).
de Paula, A. (2020). Trade networks and the strength of strong ties. In Advances in
Econometrics (Vol. 42).
de Paula, A., Richards-Shubik, S., & Tamer. E. (2018). Identifying preferences in networks
with bounded degree. Econometrica, 86(1), 263–288.
Ding, S., & Egger, P. (2020). Full-information Bayesian estimation of cross-sectional sample
selection models. In Advances in Econometrics (Vol. 42).
Dogan, O., & Taspinar, S. (2018). Bayesian inference in spatial sample selection models.
Oxford Bulletin of Economics and Statistics, 80(1), 90–121.
Elliott, M., Golub, B., & Jackson, M. O. (2014). Financial networks and contagion. American
Economic Review, 104(10), 3115–3153.
Estrada, P., & Sanchez-Aragon, L. P. (2020). Financial contagion in cross-holdings networks:
The case of Ecuador. In Advances in Econometrics (Vol. 42).
Ho, A. T. Y. (2020). Interconnectedness through the lens of consumer credit markets.
Advances in Econometrics (Vol. 42).
Jackson, M., & Watts, A. (2002). The evolution of social and economic networks. Journal of
Economic Theory, 106(2), 265–295.
Kwok, H. H. (2020). Identification methods for social interactions models with unknown
networks. In Advances in Econometrics (Vol. 42).
Liu, X., & Lee, L. F. (2010). GMM estimation of social interaction models with centrality.
Journal of Econometrics, 159, 99–115.
Liu, X. (2014). Identification and efficient estimation of simultaneous equations network
models. Journal of Business & Economic Statistics, 32(4), 516–536.
Mihoci, A., Althof, M., Chen, C. Y.-H., & Hardle, W. K. (2020). FRM financial risk meter. In
Advances in Econometrics (Vol. 42).
Paarsch, H. J., & Rust, J. (2020). Implementing Faustmann–Marshall–Pressler at scale:
Stochastic dynamic programming in space. In Advances in Econometrics (Vol. 42).
Rainone, E. (2020). Estimating spillover effects in OTC networks. In Advances in
Econometrics (Vol. 42).
Richards-Shubik, S. (2020). Application and computation of a flexible class of network
formation models. In Advances in Econometrics (Vol. 42).
Rojas, D., Estrada, J., Huynh, K. P., & Jacho-Chavez, D. T. P. (2020). Survival analysis of
banknote circulation: Fitness, network structure, and machine learning. In Advances
in Econometrics (Vol. 42).
SECTION 1
IDENTIFICATION OF NETWORK MODELS
CHAPTER 1
ABSTRACT
The authors generalize the standard linear-in-means model to
allow for multiple types with between and within-type
interactions. The authors provide a set of identification
conditions of peer effects and consider a two-stage least
squares estimation approach. Large sample properties of the
proposed estimators are derived. Their performance in finite
samples is investigated using Monte Carlo simulations.
Keywords: Networks; heterogeneous peer effects; spatial
autoregressive model; two-stage least squares; efficiency
JEL classifications: C13; C21; D62
1. INTRODUCTION
Interaction among agents can be modeled in several ways. When
the exact topology of connections is known, one possibility is to
consider the peer effects that stem from the given network
structure. There is a large and growing literature on peer effects in
economics using network data.1 A popular model employed in
empirical work is the linear-in-means model (Manski, 1993). Three
assumptions underlie this statistical model: (i) the network is
exogenous; (ii) the effects of all peers are equal; and (iii) peer status
is measured without error.
This chapter is concerned with the specification and estimation of
a peer effects model when Assumption (ii) is removed. Assumptions
(i) and (iii) are maintained.2 Specifically, we consider a model of
peer effects where different types of peers are allowed to exert a
different influence and where social interactions are different
between and within types.
We extend the conventional identification conditions of network
models (Bramoullé, Djebbari, & Fortin, 2009; Calvó-Armengol,
Patacchini, & Zenou, 2009) to multiple endogenous variables and
multiple adjacency matrices. We propose efficient two-stage least
squares (2SLS) estimators using instruments based on the multiple
reduced form approximations. We show that the standard IV
approximation (Kelejian & Prucha, 1998, 1999; Liu & Lee, 2010)
involves a very large number of IVs, even if we use a low degree
approximation of the optimal instruments.3 For this reason, we
consider many-instrument asymptotics (Bekker, 1994) allowing the
number of IVs to increase with the sample size. We also propose a
many-instrument bias-correction procedure. Simulation experiments
show that the bias-corrected estimator performs well in finite
samples. Finally, we investigate the bias occurring when the
interaction structure is misspecified. We derive analytically the bias
that occurs when only within-type peer effects are considered, that
is, when interactions between types are at work but ignored by the
econometrician. We then use a simulation experiment to evaluate
this bias in finite samples.
Our framework is a generalization of the model proposed by
Arduini, Patacchini, and Rainone (2019) to study treatment effects
with heterogeneous externalities. In their model connections are
defined by membership in a given group and all agents within the
group are connected. The availability of information on the precise
structure of interactions between agents of different types offers
alternative identification conditions based on the sparsity of
interactions within or between agent-types.4
There is a long tradition in spatial econometrics looking at spatial
autoregressive models with multiple endogenous variables (see
Kelejian & Prucha, 2004). In the spatial econometrics context,
however, the adjacency matrix is the same for all endogenous
variables. The presence of different adjacency matrices provides
alternative possibilities to identify the model that have not been
explored so far.
Our methodology is inspired by Liu and Lee (2010) and Liu
(2014). Differently from their approach where the many instruments
derive from the many networks (groups) observed in the sample, in
our model the many instruments derive from the multiple
subnetwork framework. A multiple subnetwork framework does not
only result in an increasing number of instruments but also yields
multiple approximations of the optimal instruments. We show that
the form of the many-instrument bias differs, though the leading
order remains unchanged.5
This chapter is organized as follows. The next section introduces
the econometric model. Identification conditions are given in Section
3, and in Section 4 we consider 2SLS estimation for the model.
Section 5 investigates the bias occurring when the interaction
structure is misspecified. Section 6 concludes.
where Ga,r is
formed only among nodes of type A and Gab,r keeps trace of links
from b to a.7 Regularity conditions are listed in Appendix 1. Model
(1) can be generalized in the following way:
where
and εa,r and εb,r are i.i.d. errors with variance and
respectively. Let us suppose for simplicity that If we
stack up equations (2) to (3) and restrict the endogenous effect
parameters of the two equations to be the same (i.e., ϕa = ϕb = ϕab
= ϕba), then we obtain model (1). Let us define the following
matrices
3. IDENTIFICATION
For notational convenience, from now on we omit the network
observation index. Let us define
Endogenous effects in equation (2) are identified if E(Za) has full
column rank for large n.14 In this section, we find sufficient
conditions for E(Za) to have full column rank.15 The detailed proof is
given in Appendix 3. Here, identification means that a consistent
estimator of the parameters of model (2) exists.
Proposition 1. Let Xa and Xb have full column rank. If Ma, Mb,
Ja and Jb are invertible,16 then E(Za) has full column rank in the
following cases
1. (a) i. βaϕa + γa ≠ 0,
ii. Ia, Ga and are linearly independent.
[and]
(b) i. βbϕb + γb ≠ 0,
ii. Gab and GabGb are linearly independent.
[or]
2. (a) i. γab ≠ 0,
ii. Gab and GaGab are linearly independent.
[and]
(b) i. γba ≠ 0,
ii. Ia, Ga and GabGba are linearly independent.
This implies
Hence, the approximation error by series expansion diminishes very
quickly in a geometric rate, as long as the degree of approximation
(p) increases as n increases. We can also replace Sa and Sab by a
linear combination. The instruments become
where
20
that by
Lemma B.2 in the Appendix. This converges to 0 only if the number
of instruments grows more slowly than the sample size.22 Similarly
to Liu and Lee (2010), the following corollary characterizes different
scenarios for different rates in which K diverges from n.
Corollary 1. Under Assumptions 1 –5 , (i) if
(ii) if
where
where Xa, Xb and ε ∼ N(0, 1). Borrowing from Liu and Lee (2010),
we generate the G matrix as follows. First, for the ith row of G, we
generate an integer di ε [0, 1, …, m] with a uniform probability
function, where m = 10, 20, 30. Then we set the (i + 1)th, ···, (i +
di)th elements of the ith row of G to be ones. If (i + di)th < na, the
other elements in that row are zeros; otherwise, the entries of ones
will be wrapped around such that the number of (di − na) entries of
the ith row will be ones. We partition the matrix into four
submatrices Ga, Gb, Gab and Gba with a random selection of rows
and correspondent columns. The identifier variable used to select
the two types is generated by a Bernoulli distribution with p = 0.5.
The number of replications is 1,000 and na = nb = 500. We perform
two experiments that are summarized in Tables 1 and 2. Each
column reports mean and standard error (in parenthesis) of the
empirical distributions of different estimators. The first column
shows 2SLS few IVs. It is based on equation (8) with the IV matrix
HK derived by the first-order approximation of the best instruments
(K = 24). The second column reports the 2SLS many IVs, it is
derived by the second-order approximation of the best instruments
(K = 84). Finally, the third column shows the 2SLS bias-corrected. It
is based on equation (9) with consistent estimates derived from the
2SLS few IVs.
Table 1 reports the performance of the estimators when changing
the density of the network, that is, the number of connections. Each
panel represents a different value of m, which indicates the
maximum number of connections. The data are generated with βa =
βb = γa = γb = γab = γba = 0.5. The peer effects parameters are
set to: ϕa = ϕb = 0.1 and ϕab = ϕba = 0.2. The results show that all
estimators perform well, with different nuances. In particular, one
can observe the trade-off between bias and efficiency for the 2SLS
many IVs when network density increases the higher the density, the
higher the gain in terms of efficiency with respect to the 2SLS few
IVs. However, the bias (due to the many instruments) increases as
well. The bias correction that we propose is thus particularly
beneficial when the network is dense. Table 2 reports the
performance of the estimators when changing the heterogeneity of
within and between-type parameters. The simulation setup remains
unchanged, but we now set the maximum number of connections to
20 and let the ϕ parameters vary. In the first panel, we consider ϕa
= ϕab = ϕb = ϕba = 0.1. This is the benchmark framework in which
peer effects are homogeneous. In the second panel, we introduce
some heterogeneity in the within-type interaction effects. We set ϕa
= ϕb = 0.1 and ϕab = ϕba = 0.3. In the third panel, peer effects are
different both within and between types. We set ϕa = 0.1, ϕb = 0.2
ϕab = 0.4 and ϕba = 0.05. Table 2 shows that the performance of
the estimators does not depend on the values of the parameters –
the ranking of the estimators in terms of efficiency and bias remains
unchanged.
6. CONCLUSION
This chapter extends the linear-in-means model with endogenous
and contextual peer effects by allowing them to differ within and
between types. We extend the Bramoullé et al. (2009) and Calvó-
Armengol et al. (2009) identification conditions and Liu and Lee
(2010) estimation approach when network data are available and
peer effects are heterogeneous by peer type. The bias arising when
interactions are ignored is analytically derived and evaluated in finite
samples using simulation experiments.
Note: 1,000 observations, 1,000 replications, 10 max connections, ϕb = ϕab = ϕba = 0.3,
βb = 0.5, γb = γab = γba = 0.5.
NOTES
1. See Jackson and Zenou (2014) and Jackson, Rogers, and Zenou (2017) for a collection
of recent studies.
2. For recent reviews of models and methods aiming at removing (i) and (iii) see Hsieh,
Lin, and Patacchini (2019) and Advani and Malde (2018), respectively. The approach
proposed here can also be used in conjunction with methods dealing with both (iii) and (i).
3. See Prucha (2012) for a review of Generalized Method of Moments estimators in a
spatial framework.
4. We also show that identification in this model is related to the concepts of chains and
Tree-indexed Markov chains (see Benjamini & Peres, 1994).
5. Linear-in-means models with heterogeneous peer effects may also be estimated using
Bayesian methods (see e.g., Goldsmith-Pinkham & Imbens, 2013; Hsieh & Lin, 2017). While
Bayesian methods are computationally challenging, our IV approach is of immediate
applicability for the practitioner.
6. Observe that εr generally depends upon n. Thus, the vector is a triangular array. We
suppress the subscript n for notational convenience. See Kelejian and Prucha (1998) for
details.
7. More formally,
where Ra = (Ina,r,, Ona,r,nb,r) and Rb = (Onb,r,na,r, Inb,r) are matrices that select the
nodes in types a and b, respectively. Ok,l is a k × l matrix of 0’s.
8. Ca,r is a matrix which captures all the indirect connections among nodes of type A
passing through one or more nodes of type B. Note that the ijth generic element of
Gab,rGba,r is equal to the number of length-2 paths directed from j ∈ A to i ∈ A passing
through a node l ∈ B. This matrix accounts only for distance-2 indirect connections while
Ca,r = Gab,rJb,rGba,r captures all the paths starting from j ∈ A and ending to a generic
node in B, eventually passing through other nodes of type B and finally arriving in i ∈ A
scaling them by ϕb.
9. This matrix captures all direct and indirect paths among type A nodes passing through
others type A nodes and type B nodes.
10. The assumption is crucial for identification of the model and asymptotic normality of
the estimator (see Appendix 1 for a detailed definition of a matrix uniformly bounded in
absolute value).
11. Aggregate and average models are different in terms of behavioral foundations,
contextual effects are supposed to be averages over peers in both cases w.l.o.g. see (Liu,
Patacchini, & Zenou, 2014).
12. In this case, we only have to transform equations (2)–(3), using a deviation from
group mean projector (see Liu, 2014). All the statistical results of the proposed estimator
still hold.
13. The resulting model is a SARARMAG(p; q; g) with p = 1, q = 1 and g = 2, where p
and q are the number of spatial lags for outcome and error, respectively, and g is the
number of types (see Kelejian & Prucha, 2007).
14. This implies that Assumption 4 in Appendix 1 holds.
15. Symmetric conditions and results hold for equation (3).
16. In practice we need a series expansion to approximate the inversion of the matrices.
17. For example, we can take advantage of linear independence of Ia, Ga and
(instead of Ia, Ga and ); and Gab and GaGba.
18. Note that K is a function of the degree of approximations p.
19. Our results also hold true if we observe an i.i.d. sample from a population of
networks with a stochastic but strictly exogenous structure.
20. This is a crucial assumption. See the discussion in Appendix 1 after Assumption 4.
21. To simplify the notation, we assume that n → ∞ implies na → ∞ and nb → ∞.
22. Indeed, if we use a fixed number of instruments given by the asymptotic
general. The 2SLS estimator with fixed number of instrument is generally not efficient. In
order to have efficiency, we need to index our matrix of instruments with K and let K grow
more slowly than the sample size.
23. Observe that the literature looking at spillovers on networks with missing data (point
(iii)) aims at obtaining a consistent estimator for the spillover parameter once missing data
are taken into account. Our context is different, since it allows to estimate multiple different
values for the spillover effects.
24. We use the 2SLS few IVs to ease the comparison of 2SLS estimators with the
misspecified set of instruments. Observe that the bias considered here is due to the
misspecification of the model rather than the many instrument issue.
25. The simulation results for the other cases, that is, when the maximum number of
simulations is 20 or 30, are very similar.
26. First-order approximation of optimal instruments is considered.
27. See Staiger and Stock (1997) or Baltagi, Kao, and Liu (2012) for a panel data version
of weak instrument asymptotics. Another interesting extension could be to derive the
estimator’s asymptotic properties under many weak instruments. In doing so, we are
allowing the rate of concentration parameter to be different than the rate of the sample
size. Consequently, we can compare it with the rate in which K increases. See for example,
Chao and Swanson (2005).
28. The matrices sequence is multiplied by Xa or Xb depending on the last interaction
matrix. For instance is multiplied by Xb while GbGba is multiplied by Xa.
29. In this notation a chain includes all possible paths that have common features. For
instance, all of paths are starting from A and arriving to B in the same chain.
30. Note that H1 is the IV matrix considered in Section 4, which is approximated by HK in
the feasible 2SLS estimation.
31. Note that here we need at least three chains from Ca and two from Cab because we
are considering the outcome equation for type A nodes, that is, the starting point of chains
is always type A node.
32. Given that here we are not interested in determining the transition probability law of
a chain, even if it is simple to estimate and is basically the link formation probability
considered for all of the possible combinations of nodes’ type. Benjamini and Peres (1994)
give a detailed discussion on Tree-indexed Markov chain.
33. It is equivalent to say that the probability 0 < P (gij = 1) < 1, i ∈ A, B and j ∈ A, B.
Note that transition probability can be derived from Ga, Gb, Gab and Gba. Here we are
simply excluding the classical linear in mean framework (when the matrices are complete)
and the case in which there are no connections (when the matrices are empty).
34. From a Markov Chain perspective again, a more restrictive condition consists in
assuming that the underlying Markov Chain is irreducible and aperiodic. This means that
type A is connected with type B or type A with the same probability (and the same holds for
type B). Thus, in this case tree branches with the same length have the same probability of
being observed. The aperiodicity and irreducibility are not necessary for the identification
condition to hold, but of course are sufficient.
35. Holding condition (2) instead of (1). We basically take advantage of linear
independence of Ia, Ga and GabGba instead of
36. The additional parameter restrictions (conditions (1b, 2a, or 2b) in Proposition 1) are
basically due to an additional vector in the full rank condition (i.e., E(Gabyb)).
37. Borrowing from Markov chains vocabulary again, this is because the state that
characterizes the chain is only one (A).
REFERENCES
Advani, A., & Malde, B. (2018). Credibly identifying social effects: Accounting for network
formation and measurement error. Journal of Economic Surveys, 32(4), 1016–1044.
Arduini, T., Patacchini, E., & Rainone, E. (2019). Treatment effects with heterogeneous
externalities, Journal of Business & Economic Statistics, 1–13.
https://doi.org/10.1080/07350015.019.159755
Baltagi, B. H., Kao, C., & Liu, L. (2012). On the estimation and testing of fixed effects panel
data models with weak instruments. Advances in Econometrics, 30, 199–235.
Bekker, P. A. (1994). Alternative approximations to the distributions of instrumental variable
estimators. Econometrica, 657–681.
Benjamini, I., & Peres, Y. (1994). Markov chains indexed by trees. The Annals of Probability,
22, 219–243.
Bramoullé, Y., Djebbari, H., & Fortin, B. (2009). Identification of peer effects through social
networks. Journal of Econometrics, 150, 41–55.
Calvó-Armengol, A., Patacchini, E., & Zenou, Y. (2009). Peer effects and social networks in
education. The Review of Economic Studies, 76(4), 1239–1267.
Donald, S., & Newey, W. (2001). Choosing the number of instruments. Econometrica, 69(5),
1161–1191.
Goldsmith-Pinkham, P., & Imbens, G. (2013). Social networks and the identification of peer
effects. Journal of Business and Economic Statistics, 31, 253–264.
Hsieh, C.-S., & Lin, X. (2017). Gender and racial peer effects with endogenous network
formation. Regional Science and Urban Economics, 67, 135–147.
Hsieh, C.-S., Lin, X., & Patacchini, E. (2019). Social interaction methods. In Handbook of
labor, human resources and population economics.
Jackson, M. O., Rogers, B. W., & Zenou, Y. (2017). The economic consequences of social-
network structure. Journal of Economic Literature, 55(1), 49–95.
Jackson, M. O., & Zenou, Y. (2014). Games on networks. In Handbook of game theory (Vol.
4).
Kelejian, H., & Prucha, I. R. (1998). A generalized spatial two-stage least squares procedure
for estimating a spatial autoregressive model with autoregressive disturbances. The
Journal of Real Estate Finance and Economics, 17(1), 99–121.
Kelejian, H., & Prucha, I. R. (1999). A generalized moments estimator for the
autoregressive parameter in a spatial model. International Economic Review, 40(2),
509–533.
Another random document with
no related content on Scribd:
You may laugh, children, but it is true. The dolphin had a servant,
who was also a dolphin, but of the family of the Globiceps. These are
so called because of their round heads, which look like the globes
used in the electric lighting of streets.
The young dolphin was playing in the water. He tried to attract
Pinocchio’s attention in many ways. He spouted water through the
hole which every dolphin has at the top of his head. He called to the
marionette. He smiled at the youngster. It was of no use. Pinocchio,
with his wooden nose in the air and his dough cap on one ear, would
not even turn his head.
“I wonder if he is deaf or blind?” the dolphin finally said, loudly
enough to be heard.
Pinocchio turned with a start.
“For your own benefit, I just wish to say that I am not now and never
have been deaf,” he said as haughtily as he could.
“Then why do you look at me in that fashion? And why don’t you
answer me?” was the reply.
“I am acting just as a gentleman should toward those who are
beneath him,” said Pinocchio.
“I don’t know which of us is the better of the two. All I do know is, that
my father was the richest inhabitant of the sea and that the other
dolphins considered him their king.”
“King?” mumbled Pinocchio, who knew himself to be the son of a
poor carpenter, earning so little that he never had a penny in his
pocket.
“But king or not, what does it matter? In this world we are all equal,
for we have all been created by God. Listen, my dear marionette.
Come here. As we are to travel such a long distance together, we
should be friends. Are you willing to be my friend?”
These pleasing words made Pinocchio see how stupid and how rude
he had been.
“Think of it! A fish (oh, no, I mean a sea animal) giving me lessons in
politeness!” Then turning to the dolphin, he said, “Yes, we shall be
friends. What is your name?”
“Marsovino. And yours?”
“Pinocchio.”
“A beautiful name. Come, shake hands.”
“Very willingly,” replied Pinocchio.
The good little animal stuck one of his fore fins out of the water for
Pinocchio to shake.
“And what is the tutor’s name?” said the boy of wood to the boy of
the sea.
“The tutor is a dolphin of the Tursio family, but I call him father. Is it
true that you are coming with us on our travels?”
“Yes,” said the marionette, proudly. “And I am able to teach you.”
“Teach me! That’s strange. How do you expect to teach me?”
“You will soon find out. You talk rather disrespectfully to me. I have
been in all the schools of the kingdom. And you? You probably have
never been on land for twenty-four hours.”
Marsovino looked at the marionette smilingly, but made no reply.
Pinocchio walked up and down with his hands in his pockets and his
hat at an angle of forty-five degrees, ruffling his feathers at the
brilliant remark he had made.
As soon as Tursio came near, Marsovino asked him if he were ready.
“Yes. Everything is finished,” was the reply. “Are you ready,
Pinocchio?”
“Yes. I am ready. Let us start.”
“Start? How? Do you mean to say that you are coming under the sea
with that suit?”
“Of course. It’s the only one I have.”
“A suit of paper! The very idea! Luckily I have prepared for this.
Here, Globicephalous,” he said to his servant, “give me that little suit
of ray leather,—the one I had you make this morning.”
“Splendid,” cried Pinocchio, clapping his hands. “Now I have a new
suit.”
Putting it on, he looked at himself in the water. Seeing how dark and
unbecoming it appeared, he turned to Tursio and said excitedly:
“I don’t want this. It is too ugly. I like my pretty flowered-paper one
better.”
“Your paper one Globicephalous will carry in his satchel for you.
Should you wear it in the water, it would be spoiled.”
“I want my pretty suit,” insisted Pinocchio. “If any one saw me in this
thing, he would ask me if I had been through the coal-hole.”
“But yours will be ruined if you wear it in the water, I tell you.”
“I want mine. I want mine,” wailed Pinocchio.
“Very well. Globicephalous, take the paper suit out of the traveling
bag and give it to the boy.”
The marionette turned, expecting to see an ordinary traveling bag.
Instead, he saw Globicephalous take an enormous oyster out of the
water.
“Isn’t that strange! Oyster shells for a traveling bag!”
“Strange? Why, what is strange about that?” asked Tursio.
“What is its name?” asked Pinocchio.
“That is the giant Tridacna. They are the largest oyster shells
known.”
“How large the animal inside must be,” observed Pinocchio, with a
yawn.
“Yes. It is very large, and also very beautiful. The center of the body
is a violet color dotted with black. Around this is a green border. At
the extreme edge the colors change from deepest to lightest blue.
Yes, indeed. It is very beautiful.”
“What a good meal it would make,” thought Pinocchio. His only wish
was for a good dinner, but in order to be polite he said, “Who would
ever think that there are such things under the sea!”
“Why, you have been in every school in the kingdom and don’t know
that?”
“Books on the subject you can find everywhere.”
Pinocchio bit his lips, but did not say a word. Quickly he dressed
himself again in his paper suit and declared himself ready to start.
“All right! Come along!” said the dolphin, stretching a fin out to help
Pinocchio along.
The marionette started to walk into the water. He had not gone far,
however, before his paper suit began to leave him. Hastening back
to the shore, he very meekly put on the ray-leather suit which
Globicephalous handed to him.
“Remember, my boy,” said Tursio, “that in this world of ours we must
think not only of the beauty but also of the usefulness of things. Also,
do not forget that a boy who never learns anything will never be
anything.”
“But I have learned much,” answered Pinocchio. “To prove this to
you, I can now tell you of what material this suit is made.”
“I have told you already. It is of ray leather. Do you know what a ray
is?”
“Surely I know. You may give it another name. Still, it must be that
white animal on four legs. You know. The one the shepherds shear
during some month or other.”
“Mercy!” cried Tursio. “You are talking about sheep. They give wool
to man.”
Pinocchio, without moving an eyelid, went on:
“Yes, that’s true. I have made a mistake. I should have said it is that
plant that bears round fruit, that when it opens....”
“Worse and worse,” interrupted the old dolphin. “What are you
talking of, anyway? That is the cotton plant. Marsovino, please
explain to this boy, who has read all the books in the world, what a
ray is.”
So Marsovino went on: “A ray is a fish, in shape like a large fan. It
has a very long tail, which it uses as a weapon.”
“To what class of fishes does it belong?” asked Pinocchio.
“It belongs to the same class as the lampreys, which look like
snakes, the torpedo,—”
“Be careful never to touch that fellow,” here interrupted Tursio.
“—the sawfish and the squaloids,—that is, the common shark and
the hammerhead.”
“The saw? The hammer?” observed Pinocchio. “If I find them, I must
keep them for my father. He is a carpenter, but so poor that he
seldom has money with which to buy tools.”
“Let us hope that you will never meet the saw, the terrible
hammerhead, or even the common shark,” said Tursio.
Pinocchio made no answer, but in his heart he kept thinking, “I am
very much afraid that the dolphins are teaching me, not I the
dolphins.”
Tursio then handed Pinocchio a small shell of very strange shape. It
looked like a helmet.
“Wear this, Pinocchio,” he said. “It will make a pretty cap for you.”
“It is very pretty. What is it?”
“It is a very rare shell.”
“But it is only one shell. Where is its mate?”
“It has none. It is a univalve. That means it has only one shell. The
tellines have two shells, and are therefore called bivalve. Another
kind looks like a box with a cover.”
“But does an animal live in there?”
“Of course. Every shell has its mollusk.”
“Mollusk?” repeated Pinocchio.
“Yes. The small animals that live in shells are called by that name.”
“They have a very soft body. By means of a member, called a foot,
they get such a strong hold on rocks that it is very hard to tear them
off.”
“Some mollusks have a strong golden-colored thread by which they
also hang to rocks. Why, people have even made cloth out of these
threads.”
Pinocchio cared little for all this explanation. He looked at himself in
the water, and was, after all, very much pleased with himself.
“This cap seems made for me,” he said. “Too bad I have no feather
for it.”
“Perhaps we shall find one on our journey,” laughed Tursio.
“Where will you get it? In the sea?”
“Yes, in the sea,” answered Tursio, in a tone which made the
impudent marionette almost believe him.
CHAPTER IV
“Well, children, let us hasten. If we talk so much,
the sun will rise and find us here. Come, Pinocchio!
Jump on my back and let us start.”
There was no need for Tursio to repeat his
command. In the twinkling of an eye, Pinocchio was
riding on the dolphin’s back, holding on tightly to the
dorsal fin.
“But why did the large one swallow the small one?” asked Pinocchio.
“Because the little one probably wanted to run away from the nest. It
was too soon, the little one was too young to take care of himself; so
the father took the only means he had to save the youngster from an
enemy,” patiently explained Tursio.
Just then a small fish attracted the dolphin’s attention.
“Boys,” he said, “do you see that tiny fish? It is called the pilot fish. It
is the shark’s most faithful friend. Wherever goes the shark, there
goes the pilot fish.”
“Now, Pinocchio,” he continued after a pause, “I shall leave you with
Globicephalous. Marsovino and I are going to pay a visit to the
dolphin Beluga, who is a great friend of mine. He usually lives in the
polar seas, but on account of his health, he has come to warmer
waters. We shall return this evening, if all be well. Meet us near
those two mountains which are so close together that they form a
gorge. You may take a walk with Globicephalous, but be sure to be
at that spot to-night.”
“I am ashamed to be seen with a servant,” began Pinocchio.
“You are a fine fellow,” answered Tursio, with sarcasm. “Do you
know what you should do? Buy a cloak of ignorance and a throne of
stupidity, and proclaim yourself King of False Pride of the Old and
the New World!”
With this remark Tursio turned to his pupil, and the two swam away.