Wow, this is really cool and works even better than I imagined. Thank you so much! I need to wait 12 hours to award you the bounty.

I think we are getting off track here. I was just wondering if there was a special reason for you to transpose it before using Part instead of just using Part. Even though problems can be solved in a plenitude of ways, producing the same end result, there can be subtle differences that might, objectively, make one method better than another. Doing a transpose first is a longer route, and that is why I thought you might be using it for a special reason. But I take it you are not?

data[[All,2]] -= Min[data[[All,2]]] is even faster, for small data sets such as the one above. For larger datasets (I did a RandomReal with 1 million pairs) it seems that rhermans method is winning by a bit. Nevertheless, if one wants to keep this kind of calculations fast, one should have an individual list for each column instead of combining them into a table.

for me it was 0.000057 s for the former approach and 0.000016 s for the latter. This is quite big and could certainly play a role if OP is working on larger datasets or needs to repeat the operation many times. Please do not misunderstand me, I was not looking to criticize you. I was just curious why you chose to transpose it first, if you had some special reason for it, when there is a more "straight forward" way.

BSplineCurve[] is no good due to its limited polynomial approximation of the curve. The straight lines at the beginning and end of the curve are very noticable, and if you put the same radius of the inner circle you will see that your BSplineCurve has a smaller radius.

@Martha97 I still do not understand your constraint. Do you mean that all the code for generating the arrows needs to be written within your Graphics[] function, or can you e.g. define an arrow-generating function on the lines above and then call this function inside the code above? If the latter is not possible, what is the reason? Both methods are certainly possible, but putting everything within the Graphics[] function will make it very cluttered and non-flexible.

What do you mean by applying it in the given code, and what is the reason for this constraint(?) ?

they produce the same result, but transposing the whole list first takes a lot longer time, about a factor of 4 difference on my computer. I do not know how they are implemented.

Why do you use transpose when you could simply do Min[data[[All,2]]] ?

@ihojnicki this worked, thanks! Please put it as an answer, and mention that it is available in the Options inspector under Global Options -> System configuration

Yes, it happens after any evaluation, as long as it does not finish before I manage to switch workspaces, of course. I added some more details in the question.