Computer Program PE: A Ruling from Japan
If you are growing weary of tales of AI and the impact of technology on
how we do business (and how we practise tax), I suggest that you either
move on to the next article in this issue or, perhaps, go back to the
cave from whence you came. This article considers a none-too-recent
Japanese ruling on whether algorithm-fuelled, high-frequency trading on
servers located in Japan could cause a non-resident to have a permanent
establishment (PE) in Japan.
What is high-frequency trading, anyway? Perhaps you missed Michael Lewis’s 2014
book Flash Boys,
which profiles a Canadian trader at the Royal Bank of Canada and his
discovery that high-frequency traders were using computers to gain
millisecond speed advantages over other traders. (If you missed that
book, a Google search or a ChatGPT prompt should get you caught up in
mere seconds.)
Now back to the 2011 ruling. (Yes, that’s
2011—not
a typo.) The Tokyo Stock Exchange (TSE) permits traders to set up
computer programs on servers located at sites managed by the TSE or
other firms (each firm being referred to as a “trading participant”).
Those programs place orders to buy and sell stocks on the TSE. In order
to manage an issue known as “latency” (that is, the time that elapses
between the moment a signal is sent and the moment of its receipt), it
is important that the physical distance between the server and the
stock-exchange data centre be as short as possible and, therefore, that
the servers be located in Japan.
Interestingly, the TSE requested a ruling from the National Tax Agency
of Japan on whether a non-resident conducting high-frequency trades
through the server would be considered to have a PE in Japan.
The Japanese tax authority ruled that it would not consider a PE to
exist in Japan solely on account of the specified activities, provided
that the non-resident were not entitled, at its discretion, (1) to
dispose of the server (for example, to sell, provide as collateral, or
destroy the server), or (2) to utilize and derive profit from the server
(for example, by subleasing it to third parties or by converting it to
be used for other purposes). The ruling identifies a number of
assumptions regarding the contractual arrangement between the
non-resident and the trading participant, including the following
assumptions: (1) the contract is a service agreement, (2) the trading
participant is responsible for the maintenance of the server, (3) the
non-resident is prohibited from entering the space where the server has
been installed, and (4) any restrictions imposed by the non-resident on
the trading participant with respect to the server represent the minimum
agreed-on contractual restrictions necessary to ensure the stable
provision of services and the protection of the programs, parameters,
and data on the server.
The Japanese tax authority’s determination, which closely reflects the
OECD’s and the CRA’s approaches to servers, was focused on whether the
non-resident exercises control over the server and whether the server is
at the disposal of the non-resident. Although the details of the
Japanese tax authority’s analysis are not disclosed, it appears to be
focused on comparing the server with other
physical places of business.
This ruling presents a number of interesting issues in the debate over
so-called peopleless PEs—that is, over whether a non-resident should be
considered to have a PE in a particular country in a case where the
non-resident installs technology in that country to conduct a certain
activity and would have been considered to have a PE if that technology
had been a human person. In this way, the analysis is focused more on
the autonomous functionality of the technology than on the control over
the device on which the technology operates.
With respect to computer programs installed on servers, how does an
agency analysis apply to a high-frequency-trading computer program? The
first step in the analysis is to determine whether the computer program
installed on the server could be viewed as an agent that habitually
enters into buy/sell contracts on behalf of the non-resident. The second
step would be to determine whether the program could be considered to
be an agent with independent status and thus to be excluded from being a
PE of the non-resident.
The most obvious hurdle to a finding that a computer program may be
considered an agent PE is that most agent PE definitions in the OECD’s
model tax treaties refer to a “person” acting as the agent. To my
knowledge, computer programs have not been recognized as having legal
personality. Yet.
Kim MaguireOsler Hoskin & Harcourt LLP, Vancouver