alexanu/stock2vec

Embedding stocks to vectors based on the price history

Stock2Vec

Stock prices are flutuated in every day. So, in each day, put those stocks in order of price change to one sentence. Then, with certain window size, each stock will show up with highly related stock frequently, because they tend to move their prices together.
For example, 005380(Hyundai Motors) moves together with 000270(Kia Motors). Because not only they are in same industry, but also Hyndai owns Kia.

Why it matters?

Every stock are related with others, but we couldn't represent it as a vector. With this result, we can get a similarity between stocks. Moreover, we can put those vectors to our classifier such as Deep Neural Network.

In this repo

Embedding is done with Skip-gram from Word2Vec, and GloVe, but code for embedding is not included because there are already wide spreaded.

You can figure out which stock is related with cosine similarity. In addition, some of result files are delimited with "\001" because I wanted to put those to Hive. Belows are short description about included files.

File description

• sentences.txt: Source file. Each day has one row. Each row represent a set of stocks which is sorted by price difference between the day and a day before that day.
• sentences.refined.output.txt: There are 3 columns (source stock, dest stock, similarity based on Skip-gram) and they are "\001" delimited.
• sentences.refined.output_glove.txt: There are 3 columns (source stock, dest stock, similarity based on GloVe) and they are "\001" delimited.
• sentences.refined.vectors.txt: Columns are "\t" delimited. First column represents a stock code and rest are vectors from Skip-gram.
• sentences.refined.gloves.txt: Columns are "\t" delimited. First column represents a stock code and rest are vectors from GloVe.

Sample of related stocks for 005380(Hyundai Motors)

skip-gram

code	dest_code	similarity
005380	012330	0.9679
005380	005387	0.9677
005380	005385	0.9491
005380	000270	0.9306
005380	005389	0.9176
005380	000240	0.8904
005380	009155	0.8859
005380	009150	0.8849
005380	005850	0.8805
005380	006405	0.8723
005380	034220	0.8704
005380	010690	0.8679
005380	007860	0.8672
005380	018880	0.8667
005380	086280	0.8654

glove

code	dest_code	similarity
005380	005387	0.9273
005380	000270	0.9233
005380	005385	0.9218
005380	012330	0.9100
005380	005389	0.8879
005380	033530	0.7575
005380	009150	0.7364
005380	002350	0.7042
005380	005850	0.7012
005380	018880	0.7001
005380	091180	0.6935
005380	000240	0.6934
005380	002550	0.6719
005380	003620	0.6705
005380	007860	0.6686