That depends on how much trust you have in the setup you made, versus the (financial) risk you run when it doesn’t work. I wouldn’t immediately discard the option to run it all on your own hardware, or somewhere in the cloud.
Journal entry Today's profit: $90 So the first week of hands off is complete. What are the results? 1. Live trading does not match back tests. I had a theory as to why this was the case, but was unable to confirm. 2. I realized that my algo went against my manual system which was to only trade at specific times during the day. For this week, I have modified the code to do this. There is a bug here in that it will trade multiple times in that period whereas I just want to trade once. I will sort this out later. I have also modified the code to print out more debug information during both backtests and live runs which would help me narrow down why live trading is not matching backtests. This is now active so hopefully I should have some idea in the next 24 hours. Observation: commissions take up most of the profit.
Especially if your program trades multiple times per day and you intended to trade only once. Has the bid/ask spread not an important influence as well?
Without commissions, my profit was $310 today. "The one trade" which I am aiming for generally was $800 w/o commission. If the bid/ask spread is an influence, it doesn't seem to be important (how would I know anyway?) Edit: profit was $210 w/o commissions, sorry.
I do not know what kind of products you are trading. And don't know the bid/ask spread on these. As a first estimate you could count half that spread as "extra cost" you are incurring, on top of the commission. In my case is this "half spread" often a higher value than the commission (IB's commission is not very high).
Ah OK, in that case I am trading FX and the spread seems to hover between .3 to .6 pips. And you're right, this often dwarfs the commission. What do you do about this?
Your measures to reduce the number of trades you do is already a good step to reduce the costs you incur. It also makes sense to take a critical look at how you actually execute your trades. If you are in a hurry to get your orders filled you often pay more into the spread. If you are more patient, and have more time available to fill your orders, you could create ways to pay less into the spread. But it is difficult to say in general terms how the costs of the spread can be minimized. A lot has been written about this online. One example is here: https://qoppac.blogspot.my/2014/10/the-worlds-simplest-execution-algorithim.html
Wow, that's a lot of work... I'm so Mickey Mouse right now. I actually just put a limit order on the last bar close. That seems OK, but maybe it's not?
One of the advantage of using direct API over a 3rd-party software is also data security. I have been told that anytime when you are passing data and having your algorithms executed by a third-party software, there is always a chance that your algorithms can be stolen and read by somebody. The chance of that might be small but it never hurts to be a bit paranoid. Direct API sort of eliminates that possibility. I have been told.