3 ms·

It looks like their code is assuming a fixed commission of 20 rubles, which is apparently equivalent to US$0.27. Does anyone know if that's realistic? https://g

It looks like their code is assuming a fixed commission of 20 rubles, which is apparently equivalent to US$0.27. Does anyone know if that's realistic? https://github.com/evgps/a3c_trading/blob/master/configs.py#L24 https://github.com/evgps/a3c_trading/blob/master/configs.py#... Depending on how liquid the market they studied is, code that assumes there is never any slippage may not be very realistic. There's no comparison to a simple buy-and-hold strategy, which may be less interesting from a computer science perspective but is a good way to avoid spending lots of money on transaction costs. (I once posted my own algorithmic trading project to HN that had very flawed, naive assumptions about what trades could be executed.)

hhmc 6y ago
IIRC moex is perticularly expensive to trade on, and costs can be non linear -- but something like 1bps of commms is a reasonable approximation as an upper bound. In the paper they claim a cost of 2.5RUB per transaction, not the 20 in the config file. Edit*: It looks like the comms are indeed around 10RUB per side (which is approx 1bps). https://www.moex.com/en/contract.aspx https://www.moex.com/en/contract.aspx If the model is trading single lots then this is a reasonable cost assumption, otherwise it isn't. The paper using 2.5RUB as costs is unreasonable.