🏅 Top traders

1.03

I only bought the winning bracket!

Gopfan washed 😃

@aenews He made something unbelievable strange, but what about PM? As i know you didnt buy winning bracket at all

@gonnarekt Yeah that is the joke 💀

I followed gopfan to my death.

@aenews why kalshi odds so strange? i thought u know the final result and made some trades already.

@gonnarekt Kalshi is uncertain

@aenews kalshi is ridiculously stupid, i should go there and show how to do things ^_^

@aenews I think noaa was a bigger variance from nasa than usual last month?

@ScottSupak Too lazy to check, but yeah unlike GISS there's no straightforward way to determine NOAA anomalies. So there's pretty wide variance, in general. Not surprising when some months are significantly different.

now the best estimate range is 1.01-1.03. Let’s se what we will get soon and will already push away from this.

@gonnarekt The preliminary point for the last day of ERA5 keeps July at 16.678 C.

In my ERA5->LOTI model the bias between the datasets is corrected with an adjustment of -0.04 C to 16.637 C to convert to LOTI for July (corresponding to a LOTI unadjusted of 0.973 C).

The LOTI is then adjusted by own past prediction errors for a point prediction of 0.984 C.

@parhizj it is 1.00+ already

@gonnarekt I'm running it now, finally, should be done in 40 mins or so.

For note, on the ERA5-> LOTI final data, I ended up with a final point prediction of 0.984 C after adjusting for own past prediction errors (unadjusted is 0.973 C).

edit: I setup a new python environment recently and it looks like something has gone wrong . Going to delete the old files and rerun.

edit2: third time's the charm... running again

@gonnarekt close but not quite... edit: since the release is on the 8th (and assuming they dont run it the morning of) there are only 4 more runs left.

ghcnm.v4.0.1.20250802:

99.278

ghcnm.v4.0.1.20250803:

99.171

ghcnm.v4.0.1.20250804:

100.468

oof, i lost quite a bit of mana this month

@parhizj That’s ok I lost last month cause I predicted 1.05 but it came 1.03, but after all correction they posted it 1.05 0_o

@gonnarekt I have model that predict truth, but gistemp too bad for it:)

@parhizj ghcnm.v4.0.1.20250805:

102.51

ghcnm.v4.0.1.20250805:

102.568

@parhizj 101.127 looks it settled here.

ghcnm.v4.0.1.20250806:

101.127

ghcnm.v4.0.1.20250807:

101.696

last one as the release is scheduled for 10 hours from now.

For future reference they used the ghcnm.tavg.v4.0.1.20250805 run which means they ran it on Wednesday.

this time, the (unadjusted) error was large from my ERA5->LOTI point prediction (~0.053)

I am confused why gopfan is betting so aggressively against 0.95-0.99. Probably he is taking additional data into account that provides a more precise estimate

@Weatherman he will probably end up being right again

@Weatherman 🤷 There is the GHCN daily but this is quite a bit different from the GCHN monthly dataset, although I haven't actually calculated how much. Not much motivation for me when there is no corresponding daily ERSST data to use (are you going to correlate on only the LAND data, or use some other dataset as a proxy for ERSST?)

Beyond that there is only 2 days of ERA5 data left, whatever difference of opinion more likely has to remain in the ERA5->LOTI model used:

These are the residuals that I've recorded for my own past LOTI prediction errors once all of ERA5 data is in (most of it is within the range of what might be determined by what run of ghcnm they randomly use):

At some point I started recording an extra-decimal point once I started outputting my own gistemp data that doesn't round the data since the numbers were too small:

own_data_error_residuals = np.array([-0.02, 0.01, 0.03, 0.01, 0.02, 0.03, 0.0, -0.01, 1.360 - 1.339, 1.225 - 1.173, 1.073 - 1.069, 1.032 - 1.045])

The bias is only 0.01 C....

@parhizj ERA5 data still could be off quite significantly, especially in 2024 with strong ENSO effect (El nino). The average of the daily GHCN's should give an idea how the monthly GHCN ends up though. I once looked into this data but it is a bit difficult how to process it and deal with data issues. I find this ncep reanalysis data interesting to look in as well http://www.karstenhaustein.com/climate

NCEP difference would imply 0.95 btw. Or GopFan is wrong or there is a piece of information/analysis that we're missing

@Weatherman ERA5 is a reanalysis so I don’t know whether being “off” is the right idea; maybe you mean with just different biases, uncertainty than gistemp (and its input datasets)? (This should be accounted for in the ERA5->LOTI model though). You could use either dataset for ground truth.