FUTURES

I want access to historical futures data for large swaths of the markets without having to use NinjaTrader or paying $1500/mo reeeeee

you have to go back

>>1658590
ya i did a machine learning credit rating project at school. I'm just starting in datascience though this year. I dont really know what the fuck im doing though lol

>>1658590
I got approved last year and have been trading them ever since (after 5 years in equities).
I mainly trade the /NQ and /CL, but sometimes the /ES and /SI.

Have you looked at quandl?

>>1658628
Interesting stuff.
We'll see if your 30 point /NQ drop materializes. If it does, that's a nice $600 for you. What are they making you lay out per contract, $4K?

Trades worked out
The Es vs the NQ trade they diverged but not as much as predicted. Gas I averaged into 3.57s on the overnight sell off, sold out 3.69. Not to target about 4 cents under but still a ok trade. Most of the $$$ was gas trade. The Equity models usually line up meaning go the same way, ie both positive, etc.

Thanks Computer.
Computer: beep boop beep beep

>>1658593
thinkorswim

>>1658590
I play with WTI sometimes.

I think the WTI is too high, OPEC came out last week and promissed to reduce production, but with the dollar weakening, they wont do it.

DWTI for the win boys.

>>1659916
Man WTI moves so much on rumors and news. Really tough for this algo to grab that. Learned the hard way this spring. I like trading it but what a manic product.

File: TY.jpg (1MB, 2784x1296px) Image search: [Google]

1MB, 2784x1296px

Fri
Algos only liked S&P, none of the other models cleared our thresholds.

2262
with some of the models pointing higher 2280

I bought 2247.5 and will look to add if it moves lower overnight.

>>1658599
Where at?

>>1660263
I sold out 2257
Have a good weekend

>>1658590
Nice. Can you tell me more about the startup?

>>1662136
I am the trade side, I have 2 programmers and a network engineer so a total team of 4. I traded for various market making groups in the past primarily options.

We model financial products and predict a value for the day using machine learning algos and take positions accordingly.

What I share here are the daily targets. The model also runs inter day, predicting the price the model expects over the next hour, next 30 min. So you end up with the computer throwing out predictions all day.

>>1662614
>We model financial products and predict a value for the day using machine learning algos and take positions accordingly.

how does one get into this ? where do I start ?

>>1662622
Don't buy into that bullshit. He's roleplaying. No firm has their analysts make a model for each fucking day to "trade accordingly." That would be a proprietary trading alog job that would only need one person to program and run and that person would be the nipple on the titty so to speak. THAT guy isn't lurking /biz/ thinking that /pol/ won an election or asking basic questions about buyside treasury trading (which is basic shit you would know before you got that job anyways).

OP, stop roleplaying you underage fag.

File: Phone 741.jpg (2MB, 2592x1936px) Image search: [Google]

2MB, 2592x1936px

>>1662650
But a start up would, ya know cause its a startup. Find that picture of the ten year options pit or my desk on the web.

What basic question did I ask?

>>1662711
>>1662622
can you answer my question

>>1662727
I worked in Options trading for years, gave me lots of model experience. Not only fitting the curves to the market but also managing the resulting position we had as market makers. You get to see when models work, the assumptions that are made, and most importantly how and why models fail.

The guys that wrote the actual algos were math and physics guys.

Take a look at options pricing and volatility by Natenberg. Options theory would be good intro to models. I Trade to our targets, pretty much the same way you would trade options. I have a value I think is right given the market, I defend the value in the market, buy under sell over.

>>1662760
Thanks anon

>>1662614
How sensitive are your models/trading systems to latency? Without giving away any secrets, can you share anything about the type of ML/DL you're using?

I have a background in writing hard realtime software (I write network switching fabric currently) and I've been doing some GPU-based ML/DL recently. I've been trying to build algotrading systems for a few years. The intersection of technology and economics is extremely interesting, but it's really rare to hear of small groups doing well at it.

>>1662711
Damn. That is an autistic amount of monitors.
>>>>/g/

>>1662976
Latency is not an issue at all. We do not try and compete against the HFT guys. I get our daily price targets the night before so I get plenty of time to put it on and take it off. We use our inter day predictions to more or less confirm our daily models.

On the ML/DL, we are DL. We Train our Algos over different periods of time. We run a full spectrum of models every night telling our algos to train over various time frames. So for each product I get a prediction based on the model considering the last year, 6months, 3months, 1 month of data. Depending on market conditions we choose what part of the model we should use. We have set those thresholds based on market metrics with several products influencing what model we should use. ( think VIX etc) So you prob want to compare price action over similar volatility conditions.

I will say that at this point it feels like we are looking at this the same way as some the big guys. Like it becomes self fulfilling.

>>1663131
Why you think i Love Chan

>>1658590
I'm majoring in math, planning to specialize in measure theoretic probability. What kind of probability theory/statistics techniques do you use to model these things?

>>1663758
Thanks for the reply