I wrote this one month ago, now we see this at the same time that we learn China's PBOC has ordered chinese banks to reduce dollar purchases. French yields also spiked. Murky times ahead...

This is precisely one of the settings in which good usage of LLMs can be helpful in science. See this post from @bengolub.bsky.social on the other place

En lien avec la discussion sur bsky sur attirer les chercheurs US en Europe. Pas mal likée/partagée par des collègues ayant de l'expérience en recherche en dehors de la France. Ici on a un peu de liberté académique etc, mais je doute que ça suffise!

In the other place, @pierreb.bsky.social a vu tout à fait juste, oui

Tiens, vu à l'instant, du coup je suis aussi sceptique quant à la constance de la position des insoumis sur la question...

Oui bon, LFI en est à s'inquiéter de l'image qu'on donne aux investisseurs si on saisit les actifs russes, comment dire...

Le wiki anglo l'explique clairement !
Ipp sur t^z e^-t =uv' , avec t^z=u
Donc ça devient [uv]-integrale u'v , en remarquant que v=-e^-t et u=zt^(z-1)

This was what was observed by Moll, Lions, Lasry and Gabaix in scholar.harvard.edu/files/xgabai... . These simple models don't account for the very fast increase in (income) inequality since the 80s. You need to add other assumptions that basically say some people get more lucky than others 12/x

the distribution of the distribution of the shares of wealth of each individual. This gives you something that is known as the random energy model in statistical physics, but the key point is that wealth eventually *condensates* on a few individuals, just with random luck. 6/x

I guess a very good thing to know is that wealth tends to be distributed with something called a Pareto distribution. They are remarkably stable in history and pervasive in any wealth-accumulating society. They say basically that the number of individuals with wealth >X scales as X^(-some power) 2/x

Seen in the article (but I stole the highlight from Twitter)
Europe needs to act **fast**

Well...

Tu peux aller voir, connaissant le fonctionnement de ces gens et leur amour des chambres d'écho je ne serais pas surpris que l'ajout sur ces listes se fasse de façon recursive

Discussion assez intéressante sur l'opposition que font certains entre maths/sciences vs shs, qui m'a rappelé un bout au tout début des mémoires d'Aron.
"Entre les deux langues, celle des symboles et celle des mots, il ne faut pas choisir, toutes deux contribuent à la formation de la pensée"

Of course, after this you can also try to compute what the limiting distribution is. In all generality, it's a Lévy-stable distribution, and it's a fixed point of the convolution operation, and if you derive things through convolution you can also get this scaling (and more). 10/10

So now you can think of a numerical experiment: draw N samples, compute the (finite) sum S_N, and then plot (S_N/√N)^2 as a function of N. For a power-law distribution as above, this will diverge. For A Gaussian, it doesn't, and as you expect the value of S_N^2~N. (7/10)

Well, that's when thinking about extreme values helps. What does the largest value in (X1...XN) look like? To think about this, consider the function below, which is the ccdf or survival function of X, and looks at the probability mass above a threshhold x (4/10)

Although those models can give insight in some cases, I am not sure they can capture some non-equilibrium effects. Effectively, firms do inventory management, and when you try to implement their dynamics it's very hard to avoid all sorts of feedback loops that cause endogenous instabilities

I agree with pretty much everything you've said -- but in that case shouldn't we be building *dynamic* models? This is wildly different from equilibrium network models à la Acemoglu-Carvalho
Plot from Sterman's "operational and behavioral causes of supply chain instability"