Financial Recommender Systems

Recommender Systems di
prodotti bancari-ﬁnanziari
Giovanni Semeraro , Cataldo Musto
Smart Companies and Artiﬁcial Intelligence
Firenze (Italy) - May 14, 2013

outline
• Background
• Needs Allocation
• Anima SGR’s Progettometro
• From Needs to Asset Allocation: recommender
systems
• State of the art: Collaborative filtering, content-based filtering
• Our choice: case-based reasoning
• A possible use case
G.Semeraro, C.Musto, - Recommender Systems di Prodotti Bancari-Finanziari.
Smart Companies and Artificial Intelligence, Firenze (Italy) - May 14, 2013

Background
ObjectWay Finance-as-a-Service
Smart Application Software & Services for Financial
Services Operators

Background

Background
Wealth Management reference framework

Current work
•Progettometro
• iPad app (https://itunes.apple.com/it/app/progettometro/id515222798?mt=8)
• iOs 4.3 required
• Designed by Anima SGR
• Helps people building their life projects
• Tool for needs allocation

Progettometro
proﬁle selection

Progettometro
ﬁve stereotypes

Progettometro
input information

Progettometro
projection

Progettometro
life projects

Progettometro
information about life projects

Progettometro
updated projection

Progettometro
ﬁnal report

from needs
to asset
allocation

research
question
is it possible to evolve a
needs allocation tool
towards an asset
allocation one by
exploiting artiﬁcial
inteligence techniques?

our proposal: personalization

to introduce an holistic vision of the user

to adapt asset portfolios
on the ground of personal user proﬁle and needs

to introduce a tool helpful for supporting
ﬁnancial advisors (not for private investors!)

Solution
Recommender Systems

Recommender Systems
Relevant items (movies, news, books, etc.) are suggested to
the user according to her preferences.

deﬁnition
Recommender Systems have the goal of guiding the
users in a personalized way to interesting
or useful objects in a large space of possible
options.
Burke, 2002 (*)
(*) Robin D. Burke: Hybrid Recommender
Systems: Survey and Experiments. UMUAI,
volume 12, issue 4, 331-370 (2002)

does it ﬁt our scenario?
“we are leaving the age of information, we are entering the age of recommendation”
(C.Anderson,The LongTail.Wired. October 2004)

Amazon.com
Testo
“ The technology is used by shopping websites such as Amazon, which receives about 35 percent of its revenue via product
recommendations. It is also used by coupon sites like Groupon; by travel sites to suggest flights, hotels, and rental cars; by social-
networking sites such as LinkedIn; by video sites like Netflix to recommend movies and TV shows, and by music, news, and food
sites to suggest songs, news stories, and restaurants, respectively. Even financial-services firms recently began using
recommender systems to provide alerts for investors about key market events in which they might
be interested” (N.Leavitt,“A technology that comes highly recommended” - http://tinyurl.com/d5y5hyl)

Netflix.com
Recommendations
“ The technology is used by shopping websites such as Amazon, which receives about 35 percent of its revenue via product
recommendations. It is also used by coupon sites like Groupon; by travel sites to suggest flights, hotels, and rental cars; by social-
networking sites such as LinkedIn; by video sites like Netflix to recommend movies and TV shows, and by music, news, and food
sites to suggest songs, news stories, and restaurants, respectively. Even financial-services firms recently began using
recommender systems to provide alerts for investors about key market events in which they might
be interested” (N.Leavitt,“A technology that comes highly recommended” - http://tinyurl.com/d5y5hyl)

Recommender Systems
current literature

Recommender Systems
current literature
Collaborative/Social Filtering
Content-based Filtering

collaborative recommenders
Suggest items that similar users liked in the past.

Suggest items that similar users liked in the past.
It capitalizes the
‘word of mouth’ effect

example: user-item matrix
item 1 item 2 item 3 item 4
user
1
♥ ♥
user
2
♥ ♥ ♥
user
3
♥
user
4
♥ ♥

target user: user 4
user
1
♥ ♥
user
2
♥ ♥ ♥
user
3
♥
user
4
♥ ♥

looking for like-minded users
user
1
♥ ♥
user
2
♥ ♥ ♥
user
3
♥
user
4
♥ ♥

recommendations
user
1
♥ ♥
user
2 ♥ ♥ ♥
user
3
♥
user
4
♥ ♥

content-based recommenders
Suggest items similar to those liked in the past by the user

key concepts
•Each item has to be described through a set of
textual features
•Movie plots, content of news, book summaries,
etc.

example: news recommendations
Items
♥
♥
User Proﬁle
User is
interested in
news articles
about sports,
football,
cycling, etc.

Items
♥
♥
Recommendations

Items
♥
♥
Recommendations
XG.Semeraro, C.Musto, - Recommender Systems di Prodotti Bancari-Finanziari.

Items
♥
♥
Recommendations
XG.Semeraro, C.Musto, - Recommender Systems di Prodotti Bancari-Finanziari.
Smart Companies and Artiﬁcial Intelligence, Firenze (Italy) - May 14, 2013.

both collaborative and content-based ﬁltering
are not feasible for recommending
ﬁnancial products.

CF drawback: ﬂocking

Similar users receive similar
assets.

Too many users could be moved
towards the same suggestions

consequence: price manipulation
(as in trader forums)

CBRS drawback: poor content

Features describing both assets
and private investors are very
poor (e.g. risk proﬁle)

Difﬁcult to calculate the overlap between
item and user (feature) description

Solution
Knowledge-based Recommender Systems

Knowledge-based
Recommender Systems
• Useful for complex domains
• Computers, cameras, ﬁnancial products
• Need a deep understanding of the domain
• Typically encoded by experts
• Focused on producing correct recommendations
• Focused on explanations of the recommendations

Knowledge-based
Recommender Systems
• Recommendation process
• Gets information about user needs;
• Exploits the knowledge stored in the KB to meet
user needs;
• (eventually) ask user to relax or to modify some of
the needs (e.g. expected interest rate);
• Proposes a recommendation.

we focus on a subclassof
knowledge-based recommender systems

we focus on a subclassof
knowledge-based recommender systems
case-based recommender systems

case-based RSs
• Knowledge base Case base
• Similar problems solved in the past are used as
knowledge base
• To each case is assigned a set of features
• User needs
• Description of the case
• The recommendation process consists of the
retrieval and the adaptation of similar already
solved cases

case-based RSs
solving cycle

case-based RSs
formally

case-based RSs
formally
item model

case-based RSs
formally
item model
= (model, producer, megapixel, zoom, etc.)

case-based RSs
formally
item model
= (product, asset class, macro asset class, yield, etc.)

case-based RSs
formally
item model
user model

case-based RSs
formally
item model
user model
= (risk proﬁle, experience, goals, etc.)

case-based RSs
formally
item model
user model
session model

case-based RSs
formally
item model
user model
session model
evaluation

case-based RSs
formally
item model
user model
session model
evaluation
$ 174.18
http://tinyurl.com/d3nt2fq

given a case base, it is necessary to
deﬁne similarity metrics to
compute how similar two cases are

case-based RSs
similarity

case-based RSs
similarity
state of the art:
heterogeneous euclidean overlap metric

case-based RSs
similarity
n features

case-based RSs
similarity
weight of the i-th feature

case-based RSs
similarity
distance

the retrieved solutions can be
reﬁned and modiﬁed before being
proposed to the user

solutions considered as ‘correct’
can be stored in the case base and
exploited again in the future

case-based reasoning for
ﬁnancial product recommendation

scenario
“Scrooge McDuck wants to
get richer. He decided to
invest some of his savings
and he asked for help to a
ﬁnancial advisor”

step 1
user modeling

scenario
Which features
may describe
Scrooge McDuck?

scenario
User Features
Risk Proﬁle
Financial Experience
Financial Situation
Investment Goals
Temporal Goals

scenario
User Features
Risk Proﬁle: Low
Financial Experience: High
Financial Situation:Very High
Investment Goals: Medium
Temporal Goals: Medium

scenario
User Features
Risk Proﬁle: Low
Financial Experience: High
Financial Situation:Very High
Investment Goals: Medium
Temporal Goals: Medium
MiFID-based

in a classical pipeline, the target user would have received
a “model” porfolio tailored on her proﬁle

in a pipeline fostered by a recommender system, the
ﬁnancial advisor can analyze the portfolios proposed to
similar users.

step 2
retrieval of similar users

retrieval
case base

retrieval
0.3
0.7
0.9
0.1

retrieval
0.3
0.7
0.9
0.1
similarity score

retrieval
0.3
0.7
0.9
0.1
helpful cases

in real-world scenarios, the case base contains
much more helpful cases
usually, it is necessary to introduce some strategy to diversify similar cases

case-based RSs
differentiate solutions

to each case is assigned an agreed portoﬂio
the set of the portfolios represents the set of the possible recommendations

retrieval
Obbligazionario Euro Bot 30%
Obbligazionario HighYield 15%
Obbligazionario Globale 15%
Azionario Europa 20%
Azionario Paesi Emergenti 12%
Flessibili BassaVolatilità 8%

how to combine the retrieved cases?
several strategies available

step 3
revise and review

revise and review
Obbligazionario HighYield 12.5%
Obbligazionario Globale 18.5%
Azionario Europa 21.5%
Azionario Paesi Emergenti 9.5%
rough average

revise and review
clustering (proposing diversiﬁed solutions)

ﬁnancial advisor and private investor can
further discuss the portfolio

revise and review
Original Discussed Gap
Obbligazionario
Euro Bot 30% 30%
Obbligazionario
HighYield 12.5% 10% -2.5%
Obbligazionario
Globale 18.5% 20% +1.5%
Azionario Europa 21.5% 24% +2.5%
Azionario Paesi
Emergenti 9.5% 8% -1.5%
Flessibili Bassa
Volatilità 8% 8%
interactive personalization

an evaluation score is ﬁnally assigned to the
proposed solution
yield, e.g.
retain

good solutions are stored in the case base and
exploited for future recommendations
retain

case base

(new) case base

recap
• Case-based reasoning for recommending financial products
• Goal: to help financial promoters considering solutions proposed to
similar users
• Case base: user features and agreed portfolios
• User features: risk profile (MiFID questionnaire), financial
experience, financial situation, investment goals, temporal goals
• Portfolio: model portfolio, macro asset classes, asset class
distribution, products, etc.
• Similarity: HEOM to retrieve similar ‘cases’
• Revise and Review: several strategies for cases aggregation and
combination
• Retain: considering external factors (e.g. yield) to evaluate the
effectiveness of the proposed solution

open points
• Research is not over :-)
• How to model investors?
• How to model portfolios?
• Which features should be assigned a greater weight?
• Which one is the best strategy to aggregate
recommended portfolios?
• How to model temporal constraints?
• How to consider contextual information (e.g.,
stock market situation) ?

questions?
Cataldo Musto, Ph.D
cataldo.musto@uniba.it
prof. Giovanni Semeraro
giovanni.semeraro@uniba.it

Financial Recommender Systems

Recommandé

Recommandé

Contenu connexe

Similaire à Financial Recommender Systems

Similaire à Financial Recommender Systems (20)

Plus de Cataldo Musto

Plus de Cataldo Musto (20)

Dernier

Dernier (20)

Financial Recommender Systems