1:
2: % Template article for preprint document class `elsart'
3: % with harvard style bibliographic references
4: % SP 2001/01/05
5:
6: \documentclass[doublespacing]{elsart}
7: %\documentclass[]{elsart}
8: % Use the option doublespacing or reviewcopy to obtain double line spacing
9: % \documentclass[doublespacing]{elsart}
10:
11: % the natbib package allows both number and author-year (Harvard)
12: % style referencing;
13: \usepackage{natbib, graphicx}
14:
15: % if you use PostScript figures in your article
16: % use the graphics package for simple commands
17: % \usepackage{graphics}
18: % or use the graphicx package for more complicated commands
19: % \usepackage{graphicx}
20: % or use the epsfig package if you prefer to use the old commands
21: % \usepackage{epsfig}
22:
23: % The amssymb package provides various useful mathematical symbols
24: \usepackage{amssymb}
25: \journal{Physica A}
26: \begin{document}
27:
28: \begin{frontmatter}
29:
30: % Title, authors and addresses
31:
32: % use the thanksref command within \title, \author or \address for footnotes;
33: % use the corauthref command within \author for corresponding author footnotes;
34: % use the ead command for the email address,
35: % and the form \ead[url] for the home page:
36: % \title{Title\thanksref{label1}}
37: % \thanks[label1]{}
38: % \author{Name\corauthref{cor1}\thanksref{label2}}
39: % \ead{email address}
40: % \ead[url]{home page}
41: % \thanks[label2]{}
42: % \corauth[cor1]{}
43: % \address{Address\thanksref{label3}}
44: % \thanks[label3]{}
45:
46: \title{Endo- vs. Exo-genous shocks and relaxation rates
47: in book and music "sales"}
48:
49: % use optional labels to link authors explicitly to addresses:
50: % \author[label1,label2]{}
51: % \address[label1]{}
52: % \address[label2]{}
53:
54: \author[lambi]{R. Lambiotte},
55: \corauth[cor]{Corresponding author.}
56: \ead{Renaud.Lambiotte@ulg.ac.be}
57: \author[lambi]{M. Ausloos},
58: \ead{marcel.ausloos@ulg.ac.be}
59:
60:
61: \address[lambi]{SUPRATECS, Universit\'e de Li\`ege, B5 Sart-Tilman, B-4000 Li\`ege, Belgium}
62:
63: \begin{abstract}
64: In this paper, we analyze the response of music and book sales to an external field
65: and to buyer herding. We distinguish endogenous and exogenous shocks.
66: We focus on some case studies, whose data have been collected from ranking on amazon.com.
67: We show that an ensemble of equivalent systems quantitatively respond in a similar way
68: to a similar ''external shock'', indicating roads to universality features. In contrast to
69: Sornette et al. [Phys. Rev. Lett. {\bf 93}, 228701 (2004)]
70: who seemed to find power law behaviors, in particular at long times, -
71: a law interpreted in terms
72: of an epidemic activity, we observe that
73: the relaxation process can be as well seen as an exponential one that saturates
74: toward an asymptotic state, itself different from the pre-shock state.
75: By studying an ensemble of 111 shocks, on books or records,
76: we show that exogenous and endogenous shocks
77: are discriminated by their $short-time$ behaviour: the relaxation time
78: seems to be twice shorter in endogenous shocks than in exogenous ones.
79: We interpret the finding
80: through a simple
81: thermodynamic model with a dissipative force.
82:
83: \end{abstract}
84:
85: \begin{keyword}
86: Granular models of complex systems \sep Random walks and L\'evy flights
87: \sep Self-organized systems
88: \PACS 45.70.Vn\sep 05.40.Fb\sep 05.65.+b
89:
90: \end{keyword}
91:
92: \end{frontmatter}
93:
94: % main text
95: \section{Introduction}
96:
97: The fluctuation-dissipation theorem is a corner stone of Statistical Mechanics
98: \cite{nyquist,kubo}.
99: Indeed, it allows to relate quantitatively two classes of
100: dynamical features of macroscopic systems, namely fluctuation
101: phenomena which are stochastic deviations from some equilibrium state,
102: and the dissipative response of the system to an external field.
103: In out-of-equilibrium situations, however,
104: such relations are not usually applicable {\em a priori}, even though they
105: may be generalized in some particular cases \cite{barrat,dufty}.
106: A revival of studies has been recently seen on this theorem,
107: associated to features of shocks in sociological and economical networks \cite{sornette}. Indeed, these systems are not usually considered to be mechanistic in essence, whence driven by deterministic Hamiltonian-like equations. They are not at equilibrium in the classical sense, and are usually subjects to outliers, like bubbles and crashes, in financial markets \cite{marcel}, the emergence of trends in sales and weblogs \cite{glance}, or epidemia and avalanches in opinion formation \cite{holyst}, music \cite{lam1,lam2} and science. Generally, these critical events may be caused by two kinds of mechanisms. On the one hand, there is the response to some external field. In this case, one speaks of an exogenous shock. On the other hand, there is the spontaneous evolution of the system through a hierarchy of avalanches of all sizes. These extreme events are considered to be endogenous, as it has been formalized by the theory of Self-Organized Criticality (SOC) \cite{bak}. In most non-physical systems, it is not easy to distinguish these two kinds of features since systems are usually driven by an interplay of the two mechanisms.
108:
109:
110: In the instance hereby considered, i.e. the case of marketing, the
111: sales of a product are usually driven by a reputation cascade as well
112: as advertisements for the product. For example, in book sales dynamics,
113: some books reach their peak abruptly, followed by decreasing sales, while
114: others reach their top rankings after a longer time on the market, followed
115: by gradually falling sales. Sornette et al. \cite{sornette2} introduced an
116: epidemic-like model containing a long memory process for the buyers,
117: characterized by an exponent $\theta \in[0,\frac{1}{2}]$. Endogenous shocks
118: were shown to be formed by a slow increase of the sales, followed by a symmetric
119: relaxation, i.e. the formation and the relaxation process behave
120: like $|t_c+t|^{2 \theta - 1}$, where $t=0$ corresponds
121: to the peak maximum and $t_c$ is an additional (not interpreted) positive parameter.
122: In contrast, an exogenous shock occurs abruptly and the sales decay faster,
123: like $(t_c+t)^{\theta - 1}$. This prediction agrees with the intuition that
124: an endogenous shock finds its origin in the structure of the buyers network
125: and should have a longer life time than a shock which has been imposed by
126: an external cause to the system. By measuring relaxations of a large number of
127: endogenous and exogenous
128: shocks, Sornette et al. found an average value of the exponent $\theta\sim0.3 \pm 0.1$.
129: However, the
130: values of $t_c$, i.e. the short-time behaviour of the relaxation, are not considered.
131:
132:
133: Nevertheless, despite these pioneering results, many fundamental questions
134: remain open in order to fulfill the original purpose of these studies,
135: namely a generalization of linear-response theory to marketing and sociological systems.
136: In this respect let us point to Groot \cite{groot} studies of
137: sales data in a commodity market
138: from the point of view of the fluctuation spectrum and noise correlations.
139:
140: After some caveat on
141: the methodology inherent to such a type of studies and a visual discrimination between an ''endogenous'' and an ''exogenous'' shock on a case study, in Sect. 2,
142: the present paper focuses on two important issues.
143: On the one hand, we verify a required condition for the applicability of a
144: macroscopic description, namely we check the $reproducibility$ of the ''experiment'',
145: both in music sales and on book sales in Sect. 3. This is done through two case studies.
146: Thereby, we
147: study the relaxation of equivalent systems to an external shock.
148: On the other hand, we revisit
149: the study of Sornette et al. and focus on the $short$ time scale after a
150: sales maximum ($\sim$ 1 month). On this time scale, most of the systems rather
151: show an exponential relaxation
152: that was hidden by the parameter $t_c$, and not
153: a power law decay. Moreover, we question whether the observed long time power-law
154: relaxations are not in fact associated to a saturation effect, in Sect. 4.
155: Finally in Sect. 5, we study the relaxations
156: of a ''large ensemble'' of shocks, characterized by their relaxation time $t_R$,
157: and highlight a quantitative difference
158: between exogenous
159: and endogenous shocks, whence in the short time range. The values of the
160: relaxation rates allow to discriminate rapidly between endo- and exo-genous shocks.
161: A theoretical model based on simple thermodynamics
162: taking into account a dissipative force has two easily measurable parameters given
163: by the initial and asymptotic ranking states. The relaxation time
164: has a much more precise meaning than $t_c$.
165:
166:
167: \section{Methodology}
168:
169: Amazon (www.amazon.cm)
170: is the largest online store selling many goods, such as electronic devices, books,
171: or music albums. Among its descriptions of the product, the website assigns a rank
172: which takes into account the number of copies that have been sold in the past.
173: The reverse translation of this rank into the number of sold copies is not an easy task.
174: However, as discussed in ref., one may approximate this relation by the power
175: law $S \sim R^{-1/2}$, where $S$ is the number of sold items,
176: and $R$ is its rank in the Amazon database.
177: Some warning is needed at once. The method for providing a rank to an item is
178: officially the following. For the top 10 000 best sellers from amazon.com sales.,
179: the rank is updated each hour and takes into account the sales of the preceding 24 hours.
180: The next 90 000 ranks are updated daily, while the rest of the items is updated monthly
181: with several different rules. In order to avoid such changes in the rank assignment
182: method, and therefore artificial consequences on the time evolution of $R$, we
183: restrict our analysis to items that remain in the $[1:100 000]$ interval.
184:
185: In order to get the time evolution of $R$ over a long time range,
186: we have used data collected by junglescan (www.junglescan.com).
187: This website allows the users to enter a product's URL, and scans
188: its rank $R$ from the amazon website in the course of time. The time evolution
189: of the rank is then stored and accessible. One should note, however, that the
190: scanning rate is not a constant. This is illustrated in figure \ref{smithtime},
191: for a typical case where we plot the total number of scans for some music album,
192: the $XO$ of Elliott Smith,
193: as a function of time. It is shown not only that the average scanning rate has
194: evolved in the course of time, 1/day, 3/day, 8/day, but also
195: that drastic local time interval changes may occur, as
196: illustrated in the inset. Moreover, {\it due to unrelated to our study technical problems
197: in the junglescan server}, most of the scans that we have elaborated have been stopped
198: at the beginning of November 2004.
199:
200: Let us now focus on a case study, namely {\em Angels and Demons} by Dan Brown.
201: It has been first published on July 2001 and has been scanned from Oct. 1, 2002
202: till Nov. 4, 2004. In figure \ref{browntime}, one observes that, before March 2003,
203: {\em Angels and Demons} was not under the spotlight and that its rank remained
204: near the 30 000 rank. In March 2003, there is a qualitative change obviously
205: associated with the publication and fame of {\em The Da Vinci Code}, which
206: makes the former book jumps in the top 100 after a few days. This example shows
207: the strong correlations between two book sales, namely {\em The Da Vinci Code}
208: and {\em Angels and Demons}, abrupt changes in the book sales associated to what
209: can be considered
210: an
211: $exogenous$ shock, around April 10th 2003, and an $endogenous$ shock
212: in September 2003.
213:
214: \begin{figure}
215: % Here is how to import EPS art
216:
217: \includegraphics[angle=-90,width=5.00in]{scanningRate.ps}
218:
219: \caption{\label{smithtime} Total number of scans as
220: a function of time for the music album {\em XO} of Elliott Smith. The
221: dash line corresponds to a fixed rate of 1 scan, 3 scans and 8 scans per day.
222: In the inset, we plot a zoom of this curve in an interval of 9 days in February 2002.}
223: \end{figure}
224:
225: \begin{figure}
226:
227: \includegraphics[angle=-90,width=5.00in]{daVinciRank.ps}
228:
229: \caption{\label{browntime} Time evolution of the amazon.com rank of {\em Angels and Demons}, written by Dan Brown.
230: In the inset, we zoom on the ranking during April 2003. Around September 2003,
231: there has been a slowly evolving bubble in book sales, - associated to an endogenous shock}
232: \end{figure}
233:
234: \section{Experiment reproducibility}
235:
236: \begin{figure}
237: % Here is how to import EPS art
238:
239: \includegraphics[angle=-90,width=5.00in]{macro0.ps}
240:
241: \includegraphics[angle=-90,width=5.00in]{macro1.ps}
242:
243: \caption{\label{elliott} (a) Time evolution of the rank of 3 albums of Elliot Smith.
244: The dark squares correspond to the number of newspapers citing the artist,
245: extracted from {\em www.highbeam.com}; (b) evolution
246: of the rescaled ranks of the albums in (a) around the exogenous shock. The dash
247: horizontal lines are guides
248: for the eye for the pre-shock, the shock, and the post-shock ranks }
249: \end{figure}
250:
251: A first requirement in order to
252: apply a fluctuation-dissipation theorem to sales
253: is the existence of a well-defined macroscopic friction process in the system.
254: It implies that an ensemble of equivalent systems should evolve according
255: to that macroscopic law when they are put outside the stationary state. Of course,
256: it is not easy to produce such a controlled experiment in the case of sales.
257: Nevertheless, we should verify this preliminary property through a detailed analysis
258: of statistically appropriate cases.
259:
260: \subsection{Music sales}
261:
262: In order to do so, we consider the sales of 3 albums of Elliott Smith:
263: {\em Figure 8}Ê(2000),
264: {\em XO}Ê(1998) and
265: {\em Either/Or}Ê(1998) (Fig. \ref{elliott}). On Oct. 22, 2003, this young folk music writer
266: died from an apparent suicide. The next day, in response to this unexpected focus
267: around his personality, all of his albums underwent an abrupt jump of their sales
268: that relaxed over a few weeks. This is a perfect example of exogenous shock, whose
269: accidental source moreover allows to get rid of any marketing strategy. One should
270: also note, as illustrated in figure \ref{elliott}, that the event produced a very
271: localized stream in the media, on the day of his death and the next one. This allows
272: to affirm that the slow relaxation process of the sales obviously finds its origin
273: in the sales dynamics itself.
274:
275: Let us stress that this (external) shock makes the system
276: reach a stationary state different from the pre-shock state. In the case of
277: {\em Either/Or}, for instance, the pre-shock state fluctuates around rank $1800$.
278: The shock makes the rank jump to $17$, followed by a relaxation of some weeks to
279: an asymptotic state around rank $500$. This behaviour is similar to most other
280: cases that we have studied, namely an exponential relaxation for short times
281: that saturates
282: to an asymptotic value, i.e. the post-shock state. We discuss this issue further
283: in the remainder of this section.
284:
285: The three album signals obviously differ from each other (Fig. \ref{elliott}).
286: This is well seen in the pre-shock regime where each album average rank
287: ranges between 1 800 and 8 000. A rescaling would account for emphasizing
288: the qualitative differences between the albums. Assuming a constant in
289: time scaling factor, the average pre-shock values become almost equal,
290: in so doing indicating universality features. Therefore, we observe that the three
291: data collapse on each other, and that the short time rank relaxation behaves
292: like $e^{\lambda t}$, with a relaxation coefficient
293: $\lambda=0.18$. This means that the three equivalent systems respond
294: in the same way to the studied external shock.
295:
296:
297: \subsection{Book sales}
298:
299: Another way to verify the reproducibility of an experiment is
300: to consider the response of one system to several "equivalent" shocks.
301: To do so, we have considered the rank of {\em Get with program},
302: written by B. Greene, and focused of the jumps due to his frequent
303: passages at the Oprah Winfrey Show, between Jan. 02 and Jul. 03
304: (see Fig. \ref{oprah}). The rank relaxations of the first 4 shocks are plotted
305: in figure \ref{greene}, under (a) log-normal and (b) log-log scale.
306: Data analysis leads to the same conclusion as in the previous (music sales) example, namely an initial exponential relaxation followed by saturation,
307: and confirms that book and music sales are similar macroscopic phenomena,
308: with respect to shocks.
309:
310:
311: \begin{figure}
312: % Here is how to import EPS art
313:
314: \includegraphics[angle=-90,width=5.00in]{getWith0.ps}
315:
316: \caption{\label{oprah} Time evolution of the ranks of {\em Get with the program}
317: by B. Greene }
318: \end{figure}
319:
320: \begin{figure}
321: % Here is how to import EPS art
322:
323: \includegraphics[angle=-90,width=5.00in]{get1.ps}
324:
325: \includegraphics[angle=-90,width=5.00in]{get1Log.ps}
326:
327: \caption{\label{greene}We focus on four exogenous shocks of {\em Get with the program},
328: in (a) log-normal and (b) log-log scale. The solid lines represent the exponential
329: relaxation Eq.(\ref{formula}), with $R_{\infty}=1200$, $R_{0}=6$ and $\lambda=0.16$,
330: and the power law $0.07 ~ (6+t)^{2.2}$. We also plot the short time and long time
331: asymptotic behaviours of these functions, namely $R= 6 ~ e^{0.16 t}$ and $R \sim t^{2.2}$}
332: \end{figure}
333:
334:
335:
336: \section{Short time and long time behaviours}
337:
338: \begin{figure}
339: % Here is how to import EPS art
340:
341: \includegraphics[angle=-90,width=5.00in]{long1.ps}
342:
343: \includegraphics[angle=-90,width=5.00in]{long1Log.ps}
344:
345: \caption{\label{long} We focus on the endogeneous peak of
346: {\em Heaven and Earth} by N. Roberts, in (a) log-normal and (b) log-log scale.
347: The solid lines represent the exponential relaxation \ref{formula},
348: with $R_{\infty}=3000$, $R_{0}=40$ and $\lambda=0.06$, and the power
349: law $0.045 ~ (25+t)^{2.1}$. We also plot the short time and long time
350: asymptotic behaviours of these functions for comparison, namely $R= 6 ~ e^{0.06 t}$ and $R \sim t^{2.1}$. }
351: \end{figure}
352:
353:
354:
355: The above results suggests to question the universality of
356: the power law, and to emphasize the short time scale
357: behavior relaxation, i.e. to draw a parallel
358: between the relaxation coefficient $\lambda$ and a dissipative force.
359: Moreover since
360: Sornette et al. \cite{sornette2} found relaxation processes of book sales
361: governed by power-laws $(t_c+t)^{\mu}$, it is of interest to compare
362: this power-law behaviour to the exponential one found in the previous section.
363: Notice that these authors focused their analysis on the long time behaviour. However,
364: there was no detailed analysis of the short-time scale of the relaxation in their work,
365: nor a clear explanation of the parameter $t_c$. The case studies in the previous
366: section suggest that the short-time relaxation processes are rather exponential.
367: In the following, we present a simple
368: alternative description of the relaxation that accounts
369: for the short-time dissipation, as well as for the asymptotic saturation
370: leading to the asymptotic state.
371:
372: First, let us assume that the sales can be related to the rank through the relation
373: $S=R^{- \gamma}$, $\gamma=\frac{1}{2}$. Let us stress that the exact value of
374: this exponent is not critical, and that this value was chosen only in order to allow
375: comparisons with the results of Sornette et al. Then, we assume that the response
376: of the sales to a shock is an exponential decrease toward an asymptotic state
377: different of zero.
378: The non-vanishing asymptotic state is, in the thermodynamic sense, due to the
379: continuous agitation of the buyers, agitation that encompasses internal dynamics
380: due to buyers interactions and small external kicks. The simplest form is
381:
382:
383: \begin{equation}
384: S= S_{\infty} + (S_0-S_{\infty}) e^{-\frac{\lambda}{2}t}
385: \end{equation}
386: that leads to the following expression for the item rank.
387: \begin{equation}
388: \label{formula}
389: R= (R_{\infty}^{-\frac{1}{2}} + (R_{0}^{-\frac{1}{2}}- R_{\infty}^{-\frac{1}{2}})
390: e^{-\frac{\lambda}{2}t})^{-2}
391: \end{equation}
392: This expression reduces to $R \sim R_{0} ~ e^{\lambda t}$ in the small time limit.
393: Therefore, this description has the advantage to depend directly on observable quantities,
394: namely the friction coefficient $\lambda$, and both initial and asymptotic values of the
395: ranks.
396:
397: In figures \ref{greene} and \ref{long}, we apply this fitting procedure to an exogenous
398: ({\em Get with the program})
399: and an endogenous ({\em Heaven and Earth}, by N. Roberts) shock.
400: Moreover, we compare the results with the
401: power-law fits $(t_c+t)^{\mu}$ of Sornette et al.
402: In the first case (Fig.\ref{greene}), the fitted parameters of (\ref{formula}) are $R_{\infty}=1200$, $R_{0}=6$ and $\lambda=0.16$,
403: and the power law fit is $0.07 ~ (6+t)^{2.2}$.
404: In the other case (Fig. \ref{long}), the parameters of (\ref{formula}) are $R_{\infty}=3000$, $R_{0}=40$ and $\lambda=0.06$, and
405: the power
406: law $0.045 ~ (25+t)^{2.1}$. In these figures, we also plot the short time
407: and long time
408: asymptotic behaviours of these functions for comparison.
409: One observes that both approaches lead to
410: similar results that can not be discriminated given the data accuracy. This is verified by focusing on the mean square deviation:
411: \begin{equation}
412: \sigma_F=\sqrt{\frac{\sum_{i=1}^K(\log(R^i) - \log(R_F^i))^2}{K}}
413: \end{equation}
414: where $K$ is the number of data points, $R^i$ the rank from the data set and $R^i_F$ the value of the fitted function, either exponential ($F=E$) or power law ($F=P$).
415: In the case of {\em Heaven and Earth}, for instance, these values for the exponential and the power law fit
416: are very close, $\sigma_E=0.249$ and $\sigma_P=0.278$.
417: This equivalence is remarkable, given the long time intervals considered
418: hereby ($100$ and $200$ days). However, one should note the very high
419: value of $t_c=25$ in figure \ref{long}, that dominates the power-law decrease
420: over a long time scale, and thereby might mask the extension of the
421: exponential relaxation at long times in Sornette et al. approach. It is also fair to recognize the large but similar values of the
422: exponent $\mu$.
423:
424:
425: A comparison of the friction coefficient $\lambda$ in Figs. \ref{greene} and \ref{long}
426: suggests that exogenous and endogenous shocks occur on different time scales
427: $t_R \equiv \lambda^{-1}$. One should stress here that this assertion is different
428: from that Sornette et al., i.e. a discrimination of shocks can be based on the
429: short-time behaviour of the relaxation process, and not on its tail.
430:
431:
432: \section{Universality behaviours}
433:
434: \begin{figure}
435: % Here is how to import EPS art
436:
437: \includegraphics[angle=-90,width=5.00in]{distribution2.ps}
438:
439: \caption{\label{histo} Cumulated probability of the relaxation coefficient
440: $\lambda$, in the case of exogenous and endogenous shocks. The vertical
441: lines point to the average values $0.07$ (endogenous) and $0.14$ (exogenous),
442: that correspond to the characteristic relaxation times $t_R \sim 13
443: $ and $t_R \sim 7$ days respectively}
444: \end{figure}
445:
446: In order to highlight this observation, we have focused on 111
447: (56 endo- and 55 exo-genous) shocks
448: extracted from the junglescan data, indiscriminating between books and records.
449: The shocks were found by coarse-graining the signal over one week, and by neglecting
450: relaxations with $\lambda<0.01$, such that shocks are distinguished from quick and large
451: fluctuations. Moreover, we have only considered shocks occurring at least 1 month after
452: the begin of the scans, i.e. in order to reject new products and dumping acts,
453: and whose minimum
454: rank verified $R_0<100$. Finally, exogenous and endogenous shocks were
455: visually discriminated by
456: focusing on the pre-shock acceleration.
457:
458: In order to measure $\lambda$, we have looked for the best fitting exponential
459: in an interval of 15 days after the shock. The resulting cumulated histograms for $\lambda$
460: are plotted in figure \ref{histo}. The histograms for the exogenous and the endogenous
461: shocks are markedly different. The peak and the average
462: probability distribution of $\lambda$ of the endogenous shocks are well separated
463: from those of the exogeneous ones :
464: $<\lambda>_{exo} \sim 2 <\lambda>_{endo}=0.14$.
465: This obviously confirms that the initial decay of exogenous and endogenous
466: shocks occurs on different time ranges.
467:
468: \section{Conclusion}
469:
470: We have examined endogenous and exogenous so called shocks in music and book sales,
471: measured from their rank in amazon.com.
472: We have focused on some case studies.
473: We have shown that music and book sales quantitatively respond in a similar way
474: to a similar ''external shock''. In contrast to
475: Sornette et al. \cite{sornette2}
476: who found power law behaviors, and
477: interpreted the finding in terms
478: of an epidemic activity, we have observed that
479: the relaxation can be seen as an exponential that saturates
480: toward an asymptotic state, itself different from the pre-shock state.
481: We have emphasized the non universal value of $t_c$ and found power law
482: exponents quite different from Sornette et al., both larger but very similar for
483: the two types of shocks.
484: We prefer to interpret our findings
485: through a simple macroscopic
486: model with agitated herding buyers and a dissipative force.
487: By studying an ensemble of 111 shocks, on books or records,
488: we have shown that exogenous and endogenous shocks
489: are discriminated by their $short-time$ behaviour: the relaxation time $t_R$
490: seems to be twice shorter in exogenous shocks than in endogenous ones.
491: This is a relevant (scientific and economic) result that completes
492: the discrimination procedure of
493: Sornette et al. and should be verified in other fields related to
494: trend emergence, such as opinion formation, financial bubbles or scientific avalanches,
495: on various networks
496: indicating roads to universality classes.
497:
498:
499:
500: {\bf Acknowledgements}
501:
502: Part of this work results from RL financing through the ARC
503: (02-07/293 )
504: and the CREEN (012864) project, which MA also thoroughly acknowledges.
505:
506:
507: \begin{thebibliography}{0}
508:
509: \bibitem{nyquist}
510: H. Nyquist, {\em Phys. Rev.}, {\bf 32} (1928) {110}
511:
512: \bibitem{kubo}
513: R. Kubo, {\em J. Phys. Soc. Japan}, {\bf 12} (1957) {570}
514:
515: \bibitem{barrat}
516: G. D'Anna, P. Mayor, A. Barrat, V. Loreto, F. Nori, {\em
517: Nature}, {\bf 424} (2003) 909
518:
519: \bibitem{dufty}
520: J. W. Dufty, V. Garzo, {\em J. Stat. Phys.}, {\bf 105} (2001) 723
521:
522: \bibitem{sornette}
523: D. Sornette, Monograph on extreme events, Jentsch editor (Springer, 2005)
524:
525: \bibitem{marcel}
526: M. Ausloos, P. Clippe, A. Pekalski, {\em Physica A}, {\bf 332} (2004) 394
527:
528: \bibitem{glance}
529: L. Adamic, N. Glance,
530: www.blogpulse.com/papers/2005/AdamicGlanceBlogWWW .pdf
531:
532: \bibitem{holyst}
533: J. Holyst, K. Kacperski, F. Schweitzer
534: {\em Physica A}, {\bf 285} (2000) 199
535:
536: \bibitem{lam1}
537: R. Lambiotte and M. Ausloos, physics/0508233
538:
539: \bibitem{lam2}
540: R. Lambiotte and M. Ausloos, physics/0509134
541:
542: \bibitem{bak}
543: P. Bak, C. Tang, K. Wiesenfeld {\em Phys. Rev. Lett.}, {\bf 59} (1987) {381}
544:
545: \bibitem{sornette2}
546: D. Sornette, F. Deschatres, T. Gilbert, Y. Ageon {\em Phys. Rev. Lett.},
547: {\bf 93} (2004) 228701
548:
549: \bibitem{groot}
550: R.D. Groot, L\'evy distribution and long correlation times in supermarket sales,
551: eprint arXiv:cond-mat/0412163
552:
553:
554: \end{thebibliography}
555:
556:
557: \end{document}
558: