---
title: Project16
tags: teach:MF
---

---
title: ML&F
tags: teaching
---

# ML and FinTech: Project by 尤皇倫
### key word
Machine picking
Financial risk management
return forecast
---

## 1. Motivations

---

### Motivation and why
我想藉由市場過去之歷史價格資料及其他股市之基本面之資訊分析並預測個股之未來報酬及風險，再藉由對投資者願承受風險及要求報酬，給予其投資組合建議。
近期人們強調養成投資的習慣，然而並非所有人都擁有足夠的知識自行操作，透過此程式可以快速的過濾投資者所需要的標的，給予他們除了基金和定存以外的投資選項

## 2. EDA
### Link of data
https://www.tpex.org.tw/web/index.php?l=zh-tw
https://www.twse.com.tw/zh/
https://mops.twse.com.tw/mops/web/index

I collect the data of stocks in Taiwan in 2018,2019,2020 and 2021 respectively.

### Data describtion

#### Explainatory

| No | variable | explanation |
| -------- | -------- | -------- |
| $x_{1}$ | Return | annually |
| $x_2$ | Volatility | annually |
| $x_3$ | EPS | earning per share |
| $x_4$ | P/E ratio | price/EPS |
| $x_5$ | Profit margin | Net Income/Sales |
| $x_6$ | Market value | |

#### output
Expect return for 1 year
Expect volitility for 1 year

### Missing value
2018
2019
2020

### disturbution of return and volitility
2018

2019

2020

2021

### Correlation heatmap
2018

2019

2020

## 3. Problem formulation
#### Model setting
Tesorflow:Keras
Loss function: MSE
Optimizer: Adam
Activation function: Relu

#### Benchmark model

We consider a pool model with muliple linear regression:

$$r_{t+1} = \beta_0+\beta_1r_{t}+\beta_2\sigma_{t}+\beta_3 x_{3,t}+\beta_4 x_{4,t} +\beta_5 x_{5,t}+\varepsilon_{t},\;\;t=1, 2$$
$$\sigma_{t+1} = \beta_0+\beta_1r_{t}+\beta_2\sigma_{t}+\beta_3 x_{3,t}+\beta_4 x_{4,t} +\beta_5 x_{5,t}+\varepsilon_{t},\;\;t=1, 2$$

#### Process
We first use the data in 2018, and return and volitility in 2019 to trian the model by ANN and LR respectively

Performance of ANN model I

Performance of LR model I(benchmark)

Then we use the data in 2019, and return and volitility in 2020 to trian the model by ANN and LR respectively

Performance of ANN model II

Performance of LR model II(benchmark)

### Backtest
then we use the model to predict the result and compare with the actual result as backtest
Performance of ANN model I

Performance of MLR model I(benchmark)

Result of ANN model I

Result of MLR model I(benchmark)

Actual result I

Performance of ANN model II

Performance of LR model II(benchmark)

Result of ANN model II

Result of LR model II(benchmark)

Actual result II

Build portfolio

I use the predict result to build portfolio by models respectively.

I choose the stocks that
volatility < mean
return > mean


ANN model I
return=0.166
mean of volatility=0.4123
portfolio volatility=0.16682866

LR model I(benchmark)
return=0.054
mean of volatility=0.4532
portfolio volatility=0.0793308

ANN model II
return=0.24
mean of volatility=0.369
portfolio volatility=0.12767422

LR model II(benchmark)

return=-0.029
mean of volatility=0.282
portfolio volatility=0.12128712

## 4. Analysis and Conclusion
1.The MSE of ANN model is not outperform than LR model.
2.The portfolio built by ANN has higher return than LR model.
3.The portfolio built by ANN has higher volatility than LR model however with the effect of diversification the portfolio volatility would decrease.

## Reference

---

---