Wrong results on clean trade record dataset

[monissa-blake]

Hi, here are my trade records:

        dates    amounts
0  2008-01-21 -26.201664
1  2008-01-22  28.204748
2  2008-10-10 -25.609337
3  2008-10-13  27.471967
4  2008-12-17 -15.539259
5  2008-12-29  12.937634
6  2009-01-19 -11.579093
7  2009-01-22  10.831304
8  2009-02-02 -11.577097
9  2009-02-04  12.156783
10 2011-09-12 -13.066551
11 2011-09-13  13.246472
12 2011-11-01 -14.440242
13 2011-11-03  14.763795
14 2016-06-24 -24.641554
15 2016-06-29  22.719878
16 2020-03-09 -23.999442
17 2020-03-10  24.802491
18 2020-03-12 -22.086328
19 2020-03-13  23.743891
20 2020-03-16 -19.685434
21 2020-03-20  19.447014
22 2025-10-20 -67.723413
23 2025-10-21  67.593747

When I run this:

try:
    rate = pyxirr.xirr(df)
except Exception:
    rate = np.nan

print (rate)

I don't get any error and the output is 475962695445.82764 which is not the annual return.
I guess the numerical solvers may diverge, but I don't know why.
Is it a bug, or pyxirr can't deal with such data?

[Anexen]

@monissa-blake, thanks for submitting the issue.

You're absolutely right - when the root-finding algorithm diverges, XIRR can return meaningless results. I'll look into improving the stability and handling edge cases in future releases.

That said, XIRR doesn't make much sense for your particular data. Your series looks like a set of short, paired trades (each outflow followed by an inflow, often returning to near zero), so no single internal rate of return meaningfully summarizes them.

A more appropriate way to summarize performance is to compute per-trade returns and analyze them using metrics such as:

• Average or median trade return
• Win rate
• Profit factor

If you want to evaluate hypothetical portfolio-level performance, you could construct a cumulative portfolio value and then compute metrics like:

• Compound Annual Growth Rate (CAGR) - note this assumes reinvestment of trade results and is therefore a hypothetical measure in this context.
• Time-Weighted Return (TWR) - meaningful only if applied to a constructed portfolio value series, not raw individual trades.