Page 101 - Special Topic Session (STS) - Volume 4

STS566 K. Prokopenko et al.
Banks can use this input for the long-term planning for cash printing and
destruction.

2. Data research
Analysis was provided using more than 3 years of daily total cash payments
from 670 branches of a wide cash payments network in Western Europe. From
the data processing perspective, cash payments (deposits and withdrawals)
can be interpreted as daily time series which may contain trend and
seasonality parts. It was highly important to analyze and understand the
structure of the data and to find any regularities which should be used for the
further modeling. The size of the daily-aggregated data sample is about 1200
points for incoming payments and 1200 points of outgoing payments for each
branch.
Global trend and annual seasonality. Linear regression approach [2] was
used for linear annual trend estimating. Autocorrelation function (ACF) was
chosen as a tool for data structure analysis during research on weekly-
aggregated data samples. There are strict spikes on 53th points which proves
annual 52 weeks seasonality for both incoming and outgoing payments. The
annual seasonality is more intensive for outgoing payments instead of less
intensive annual and strict 4-5 weeks seasonality for incoming payments
(Fig.1).

90 | I S I W S C 2 0 1 9