affixalex · September 22, 2023 04:58
diff --git a/gistfile1.txt b/gistfile1.txt
 This is a strategy intended to be executed at around 8AM ET through the first 
 hour of trading (10:30AM ET). It's a refinement of my previous approach aimed 
 at generating automated buy signals at market open. Note that I didn't actually
 write the machine learning model for sentiment analysis, it's a black box to me,
 but it performed very well on the training dataset. I'm sure there are better 
 proprietary sentiment models floating around.

 1. Find after hours gappers, big movers from previous close to current open
 2. Only select gappers priced from $10-20 with a float of 1M+ shares
 3. The top 50 gapping symbols form the watch list for this period
 4. Use the newsfilter.io streaming API to filter for news on these symbols
 5. Scrape source URL (separate async queue) to perform sentiment analysis
 6. Average of sentiment over 5+ articles forms strong signal for scalp entry
 7. Consider 20% equity stake on stop 5 gappers with 75+% positive sentiment
 8. Stop loss and take profit at 2:1 ratio, risking $2500 to make $5000

 A single trade entry and exit is a somewhat naive approach, as it doesn't give 
 any chance to catch heroic runs. Selling 25% of the position at 2:1 target
 price is probably preferable here. The Alpaca bracket order API doesn't seem to
 facilitate this in one request, so it'd probably require selling the whole 
 position and buying back with the proceeds. This would get a little complex in 
 terms of the implementation, probably recursing into the same code path to just
 place new bracket orders on the way up with some decay model.

 Also, this approach will require running on some cloud provider both to fetch 
 the articles and perform the sentiment inference. AWS Data Pipeline?

 References:

 https://newsfilter.io/api-plans

 https://developers.newsfilter.io/

 https://huggingface.co/mrm8488/distilroberta-finetuned-financial-news-sentiment-analysis/tree/main
	This is a strategy intended to be executed at around 8AM ET through the first
	hour of trading (10:30AM ET). It's a refinement of my previous approach aimed
	at generating automated buy signals at market open. Note that I didn't actually
	write the machine learning model for sentiment analysis, it's a black box to me,
	but it performed very well on the training dataset. I'm sure there are better
	proprietary sentiment models floating around.

	1. Find after hours gappers, big movers from previous close to current open
	2. Only select gappers priced from $10-20 with a float of 1M+ shares
	3. The top 50 gapping symbols form the watch list for this period
	4. Use the newsfilter.io streaming API to filter for news on these symbols
	5. Scrape source URL (separate async queue) to perform sentiment analysis
	6. Average of sentiment over 5+ articles forms strong signal for scalp entry
	7. Consider 20% equity stake on stop 5 gappers with 75+% positive sentiment
	8. Stop loss and take profit at 2:1 ratio, risking $2500 to make $5000

	A single trade entry and exit is a somewhat naive approach, as it doesn't give
	any chance to catch heroic runs. Selling 25% of the position at 2:1 target
	price is probably preferable here. The Alpaca bracket order API doesn't seem to
	facilitate this in one request, so it'd probably require selling the whole
	position and buying back with the proceeds. This would get a little complex in
	terms of the implementation, probably recursing into the same code path to just
	place new bracket orders on the way up with some decay model.

	Also, this approach will require running on some cloud provider both to fetch
	the articles and perform the sentiment inference. AWS Data Pipeline?

	References:

	https://newsfilter.io/api-plans

	https://developers.newsfilter.io/

	https://huggingface.co/mrm8488/distilroberta-finetuned-financial-news-sentiment-analysis/tree/main
No results found