Transforming Data Streams with Kafka Connect: An Introduction to Single Messa...
CatchSense
1.
2. The
Problem
• Current
forecasting
methods
are
arbitrary and
not
rigorously
back-‐tested.
• Rear-‐view looking
data
is
used;
does
NOT
reflect
forward
looking
realities.
• Real
Estate
is
a
fundamentals
(demand
and
supply)
driven
industry
with
huge
sums
at
stake.
• Risk/Returncan
be
optimized with
data-‐driven
information
and
insights.
3. What
we
do
Using
Deep
Learning
methods,
we
process
more
than
20
years
worth
of
Macroeconomic
and
capital
market
data,
historical
transactions,
and
property
features
to
forecast
near-‐term
price
trends
in
Singapore,
enabling
market
players
to
make
better-‐informed
decisions.
4. Data
à Insights
Unit
psf
Macroeconomic
e.g.
GDP,
Stock
Index,
Housing
Supply,
URA
Property Price
Index
Past
transactions
Official
URA
data with
outliers
removed
Property
features
e.g. Distance
to
MRT,
schools,
floor,
area,
tenure
Data
Machine
Learning
Algorithms
Insights
Feeback
5. URA
Property
Price
Index
Project
average
psf
Unit
psf
The
Development
Process
6. Forecasting
the
URA
Property
Price
Index
The
red
segment
represents
the
results
forecasted by
our
algorithm.
7. Measuring
Accuracy
-‐ Mean
Absolute
Percentage
Error
(MAPE)
Where At is
the
actual
value
and Ft is
the
forecast
value,
the
difference
between At and Ft is
divided
by
the
Actual
value At again.
The
absolute
value
in
this
calculation
is
summed
for
every
forecasted
point
in
time
and
divided
by
the
number
of
fitted
points n.
Multiplying
by
100
makes
it
a
percentage
error.
8. Percentile
(%) MAPE
(%)
50.0 4.3
75.0 8.1
80.0 9.3
Error
distribution
– Average
Project
psf
9. How
do
we
compare?
MAPE
*Zillow (%) CatchSense (%)
Within 5% 53.9 55.8
Within 10% 75.6 81.7
Within
20% 89.7 95.3
Median 4.5 4.3
*Zillow
is
the
leading
property
portal
in
the
US,
listed
on
the
NASDAQ
with
a
$4.5b
market
cap.
10. Error
by
region
Region *MdAPE (%)
Core Central
Region 4.9
Outside
Central
Region 3.8
Rest
of
Central
Region 4.0
Median 4.3
*MdAPE uses
Median
Absolute
Percentage
Error
instead
of
Mean
13. In
the
Pipeline
• Extracting
drivers
(coefficients)
of
price
movements
• Displaying
comparable
transactions
to
support
forecasts
• Investment
tool
to
rank
projects/units
by
appreciation
potential
• Applying
model
to
commercial
property
and
rentals
14. Why
CatchSense?
• A
scientific,
data-‐driven
forecasting
tool
for
pricing
• Make
better-‐informed
decisions
before
investing/divesting
• Independent,
trusted
third-‐party
projections
15. We
use
data
to
answer
these
questions
❔ When
to
invest/divest
– emerging
trends,
turning
points,
risks
❔ Where
to
invest?
❔ What
to
build
– winning
product
attributes
❔ Who
to
sell
to?
❔ How
much
to
charge?