Archivio istituzionale della Ricerca - Bocconi
Not a member yet
30225 research outputs found
Sort by
Random features Hopfield networks generalize retrieval to previously unseen examples
It has been recently shown that a feature-learning transition happens when a Hopfield Network stores examples generated as superpositions of random features, where new attractors corresponding to such features appear in the model. In this work we reveal that the network also develops attractors corresponding to previously unseen examples generated as mixtures from the same set of features. We explain this surprising behaviour in terms of spurious states of the learned features: increasing the number of stored examples beyond the feature-learning transition, the model also learns to mix the features to represent both stored and previously unseen examples. We support this claim by computing the phase diagram of the model and matching the numerical results with the spinodal lines of mixed spurious states
Phase diagram of compressed sensing with l0-norm regularization
Noiseless compressive sensing is a two-step setting that allows for undersampling a sparse signal and then reconstructing it without loss of information. The LASSO algorithm, based on l1 regularization, provides an efficient and robust way to address this problem, but it fails in the regime of very high compression rate. Here we present two algorithms, based on l0 -norm regularization, that outperform LASSO in terms of compression rate in the Gaussian design setting for measurement matrix. These algorithms are based on the approximate survey propagation, an algorithmic family within the approximate message Passing class. In the large system limit, they can be rigorously tracked through state evolution equations, and it is possible to exactly predict the range compression rates for which perfect signal reconstruction is possible. We also provide a statistical physics analysis of the l0 -norm noiseless compressive sensing model. We show the existence of both a replica symmetric state and a one-step replica symmmetry broken (1RSB) state for sufficiently low l0 -norm regularization. The recovery limits of our algorithms are linked to the behavior of the 1RSB solution
Taxes today, benefits tomorrow
This paper tests whether partially unemployed workers value future preserved benefits when they bunch at the kink of the unemployment insurance benefit-withdrawal schedule. I extend the bunching formula of Saez (2010) to a dynamic setting that accounts for the value of future benefits tied to taxation. This yields new tests of tax-benefit linkage based on bunching heterogeneity. I verify in quasi-experiments that UI extension programs that decrease the value of future preserved benefits lead to more bunching and to lower labor supply. Last, a quantification exercise of the dynamic bunching formula provides extra support for a strong tax-benefit linkage
Ratings and banking regulation: a shift from productive (Basel II), to contradictory (EBA-GL LOM and supervisory Practices), to dangerous (Basel 3+ and CRR3)
External ratings (agency ratings) and internal ratings were already in use before Basel II, as powerful management tools. Basel II "adopts" them (the former in the Standard approach, the latter in the IRB approaches) in a productive way: internal ratings represent the final summary of the creditworthiness assessment of debtors/transactions, to be used both in lending/review processes and in those aimed at regulatory capital adequacy measures, to be developed with a medium to long-term target horizon, leaving banks with the discretion to choose the assignment methods, as long as all relevant information (including qualitative and forward-looking data) is included in the judgment.
Instead, in the EBA-GL LOM and supervisory practices, internal ratings are predominantly the result of statistical tools in which the role of behavioural information is much broader compared to that of qualitative, strategic, and prospective information; they target short-term forecasting horizons, approaching early warning systems used in ongoing monitoring; and they are often definitively approved by bank officers different from those who underwrite the loans.
Basel 3+ confirms the Basel II framework, ignores the contradictions in the EBA-GL LOM and supervisory practices, and generates the false perception that internal ratings are no longer essential. In addition, CRR3 emphasizes the use of external ratings also in the SMEs segment and encourages the creation of new rating agencies (including those linked to central banks), thus pushing toward a commodity-oriented logic in bank-SMEs relationships (in contrast with the EBA-GL LOM).
One may wonder whether the evolution of the role and content of ratings in regulations and supervisory practices is the result of a deliberate, conscious shift in approach or if it is simply the outcome of the occasional predominance of differing positions, without any central authority, even a "Czar of ratings.
Who care about childcare? Covid-19 and gender differences in local public spending
The Covid-19 pandemic increased the salience of childcare and focused attention
on the allocation of public funds for it. Focusing on Italy, one of the first countries
severely impacted by the Covid-19 crisis, we analyze how male and female politicians
responded to the Covid-19 pandemic in the allocation of funds to childcare.
To assess causality, we analyze close mixed-gender races in Italian local elections in
small municipalities without gender quotas from 2016 to 2023. Our findings show
that pre Covid-19 female mayors spent more on childcare than male mayors. However,
during the pandemic, the gender gap closed, as male mayors increased spending,
a trend that continued in the post pandemic period
Decisori ed emergenze. Brevi considerazioni introduttive
Lo scritto introduce la sezione monografica “Decisori ed emergenze” della rivista DPCE Online. La sezione monografica è dedicata allo studio dell’emergere di nuovi decision-makers, diversi dal decisore politico tradizionale, nel corso degli ultimi venticinque anni caratterizzati da crisi di diversa natura (terroristiche, pandemiche, belliche, finanziarie)
Green grease: environmental crime, corruption and money laundering
We offer a theoretical and empirical analysis that shows how corruption can be a ‘green grease’ that facilitates both environmental crimes and money laundering. Corruption can be an accelerator that amplifies the multiplier effect on illegal revenues. We
model these relationships using a logistic function, which we calibrate to estimate overall environmental crime revenues using
Italian regional data from 1995 to 2020
Difendere per conciliare: il nuovo volto del patrocinio nel contesto della giustizia conciliativa
Questo contributo si propone di analizzare alcuni aspetti legati alla rappresentanza tecnica nel contesto del tentativo di conciliazione obbligatorio (re) introdotto dall’ultima riforma del processo civile (c.d. Riforma Cartabia) e di quello su richiesta delle parti, di cui all’art. 185 c.p.c. Ai sensi del riformato art. 183, comma 3, c.p.c.: «il giudice interroga liberamente le parti, richiedendo, sulla base dei fatti allegati, i chiarimenti necessari e
tenta la conciliazione a norma dell’articolo 185». In questo nuovo quadro normativo, l’avvocato assume un ruolo cruciale sia nel tentativo di conciliazione obbligatorio di cui all’art. 183 c.p.c. che in quello su richiesta delle parti previsto dall’art. 185 c.p.c. La rappresentanza tecnica con riguardo alle attività da svolgere per giungere a un risultato conciliativo in seno al processo pone due principali ordini di questioni che saranno oggetto in questo contributo: le prime, con riferimento ad alcuni aspetti tecnici del ruolo della difesa sorti alla luce del nuovo impianto normativo; le seconde, con riferimento ai nuovi doveri dell’avvocato
Pandemic consumption
This paper examines how households adjusted their consumption behavior in response to COVID-19 infection risk during the early phase of the pandemic and without consumption lockdowns. We use a monthly consumption survey specifically designed by the German Statistical Office, covering the second wave of COVID-19 infections from September to November 2020. Households reduced their consumption expenditures on durable goods and social activities by 24 percent and 36 percent, respectively, in response to one hundred additional infections per one hundred thousand inhabitants per week. The effect was concentrated among the elderly, whose mortality risk from COVID-19 infection was arguably the highest