{ "id": 581093, "title": "Kaggle v.s. Real-quant Problem", "url": "/competitions/drw-crypto-market-prediction/discussion/581093", "author_name": "", "post_date": "2025-05-28T09:50:38.192912800Z", "votes": 2, "comment_count": 4, "views": 0, "content": "

Wondering any full-time or intern quant researchers in this competition?

\n

Clearly, the given freq is 1 min in crypto market, I assume, the latency in crypto is milli-sec level, after colo aws server in Tokyo?

\n

The major choice in this project is Ridge ensemble on LGB + XGB over 800+ features ( could be more if user-designed features added)

\n

Under this scenario, can the Tick-to-Order latency satisfy the real trading speed?

", "messages": [ { "id": "3211314", "postDate": "05/28/2025 09:50:38", "content": "

Wondering any full-time or intern quant researchers in this competition?

\n

Clearly, the given freq is 1 min in crypto market, I assume, the latency in crypto is milli-sec level, after colo aws server in Tokyo?

\n

The major choice in this project is Ridge ensemble on LGB + XGB over 800+ features ( could be more if user-designed features added)

\n

Under this scenario, can the Tick-to-Order latency satisfy the real trading speed?

", "rawMarkdown": "Wondering any full-time or intern quant researchers in this competition?\n\nClearly, the given freq is 1 min in crypto market, I assume, the latency in crypto is milli-sec level, after colo aws server in Tokyo?\n\nThe major choice in this project is Ridge ensemble on LGB + XGB over 800+ features ( could be more if user-designed features added)\n\nUnder this scenario, can the Tick-to-Order latency satisfy the real trading speed?", "votes": null }, { "id": "3211323", "postDate": "05/28/2025 09:55:03", "content": "

especially in market taker strategy, as we are predicting price movement.

\n

if tick-to-order latency is large, significant alpha CANNOT be caught, as taker order cannot be fiilled or with HUGE slippage.

", "rawMarkdown": "especially in market taker strategy, as we are predicting price movement. \n\nif tick-to-order latency is large, significant alpha CANNOT be caught, as taker order cannot be fiilled or with HUGE slippage.", "votes": null }, { "id": "3211353", "postDate": "05/28/2025 10:33:36", "content": "

Although the given freq is 1 min here, the frequency of label that we are to predict seems relatively long-term. I estimate it's 1 hour or longer.

", "rawMarkdown": "Although the given freq is 1 min here, the frequency of label that we are to predict seems relatively long-term. I estimate it's 1 hour or longer.", "votes": null }, { "id": "3211480", "postDate": "05/28/2025 13:32:53", "content": "

cool, how you get the sense on >=1 hour ?

", "rawMarkdown": "cool, how you get the sense on >=1 hour ?", "votes": null }, { "id": "3211537", "postDate": "05/28/2025 14:28:29", "content": "

You can have a look at the ACF/PACF of the label. The ACF tails off slowly and PACF cuts off at lag1. So the label should be long-term price movement. And you can compare the value with the price movement of BTC. It should be around 1 hour.

", "rawMarkdown": "You can have a look at the ACF/PACF of the label. The ACF tails off slowly and PACF cuts off at lag1. So the label should be long-term price movement. And you can compare the value with the price movement of BTC. It should be around 1 hour.", "votes": null } ], "comments": [ { "id": 3211323, "author_name": "henrysun", "author_url": "", "post_date": "05/28/2025 09:55:03", "content": "

especially in market taker strategy, as we are predicting price movement.

\n

if tick-to-order latency is large, significant alpha CANNOT be caught, as taker order cannot be fiilled or with HUGE slippage.

", "votes": null, "replies": [] }, { "id": 3211353, "author_name": "yw2735", "author_url": "", "post_date": "05/28/2025 10:33:36", "content": "

Although the given freq is 1 min here, the frequency of label that we are to predict seems relatively long-term. I estimate it's 1 hour or longer.

", "votes": null, "replies": [ { "id": 3211480, "author_name": "henrysun", "author_url": "", "post_date": "05/28/2025 13:32:53", "content": "

cool, how you get the sense on >=1 hour ?

", "votes": null, "replies": [ { "id": 3211537, "author_name": "yw2735", "author_url": "", "post_date": "05/28/2025 14:28:29", "content": "

You can have a look at the ACF/PACF of the label. The ACF tails off slowly and PACF cuts off at lag1. So the label should be long-term price movement. And you can compare the value with the price movement of BTC. It should be around 1 hour.

", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3211314": "Wondering any full-time or intern quant researchers in this competition?\n\nClearly, the given freq is 1 min in crypto market, I assume, the latency in crypto is milli-sec level, after colo aws server in Tokyo?\n\nThe major choice in this project is Ridge ensemble on LGB + XGB over 800+ features ( could be more if user-designed features added)\n\nUnder this scenario, can the Tick-to-Order latency satisfy the real trading speed?", "3211323": "especially in market taker strategy, as we are predicting price movement. \n\nif tick-to-order latency is large, significant alpha CANNOT be caught, as taker order cannot be fiilled or with HUGE slippage.", "3211353": "Although the given freq is 1 min here, the frequency of label that we are to predict seems relatively long-term. I estimate it's 1 hour or longer.", "3211480": "cool, how you get the sense on >=1 hour ?", "3211537": "You can have a look at the ACF/PACF of the label. The ACF tails off slowly and PACF cuts off at lag1. So the label should be long-term price movement. And you can compare the value with the price movement of BTC. It should be around 1 hour." }, "source": "meta" }