README.md

CTA_fundamental_strategy

This markdown file is the framework to demonstrate the research process of a fundamental strategy in China commodities. It also shows part of the backtesting results in the metal sector, including copper, aluminium, zinc and nickle.

Table of Contents

• I. Overview of Data
• II. Backtesting Philosophy
• III. Backtesting Results
• IV. Return Models
• V. Risk Models
• VI. Future Work

I. Overview of Data

The data source comes from a database merchant called Mysteel. We selected fundamental data on daily and weekly frequency, which have more than five years‘ historical data. The raw data can be classified in four kinds:

• Price: including future prices, spot prices, cost of goods, profits of goods
• Supply: including the productivity effenciency, supply quantity
• Demand: including the trading volume
• Inventory: including the social inventory and exchange‘s warehouse inventory

II. Backtesting Philosophy

Step 1: Data Preparation

• Check the data quality:
  • Detect the outliers
  • Check the delay frequency of the publish time for the historical data
• Data cleaning:
  • Remove the outliers
  • Trace the real publish time

Step 2: Construct the Return Models for each commodity

• Construct Factors:
  • POP(period over period)
  • YOY(year over year)
  • Momentum
  • Z-score, Quantile
• Calculate Factor Returns in the historical window from 2015 to 2020.
• Explore effective factors that is relative to the daily return of commodities from both the macro-economics and quantitative prospective.
• Form the factor infrastructure as follows. There are three layers in the factor architecture.
  • The top layer is the style of fundamental data, in the aspects of prices, supply, demand and inventory.
  • The middle layer is the data subjects for each style.
  • The bottom layer is the derived factors of data subjects which are effective in our analysis.

• Factor Combination by equal-weighting in each layer. Generate signals to predict daily return for each commodity.

Step 3: Construct the Risk Model for the portfolio

We used Target Volatility Strategy to tune weights on each commodity. The strategy works as follows:

1. First consider Target Volatility for Single commodity. If we set a fixed target volatility for each commodity by leveraging, then the weight of i-th commodity would be $$ w_{t+1}^i = \frac{\sigma_{tgt}}{N\sigma_t^i}$$
2. Second consider the correlation among commodities. The volatility of the portfolio is $$\sigma_p = \sqrt{\sum_{i=1}^Nw_i^2\sigma_i^2 + 2\sum_{i=1}^{N}\sum_{j=i+1}^Nw_iw_j\sigma_i\sigma_j\rho_{ij}}$$
3. If the weights of all commodities are adjusted to target volatility, the volatility of the portfolio is $$ \sigma_P=\frac{\sigma_{tgt}}{N}\sqrt{N+2\sum_{i=1}^N\sum_{j=i+1}^N\rho_{ij}} $$
4. The average of correlation is: $$ \bar{\rho}= \frac{2\sum^N_{i=1}\sum^N_{j=i+1}\rho_{ij}}{N(N-1)}$$
5. The equation of volatility becomes: $$ \sigma_P = \sigma_{tgt}\sqrt{\frac{1+(N-1)\bar{\rho}}{N}} $$ It shows that the diversification makes the volatility of portfolio lower than the volatility of single commodity.
6. Adjust the target volatility of single commodity according to the correlation factors: $$ \sigma_{tgt} = \sigma_{P, tgt}\sqrt{\frac{N}{1+(N-1)\bar{\rho}}} = \sigma_{P, tgt} \times CF(\bar{\rho})$$
7. Tune the weight of each commodity by $$ w_{t+1}^i = \frac{\sigma_{P,tgt}}{N\sigma_t^i}\sqrt{\frac{N}{1+(N-1)\bar{\rho}}} = \frac{\sigma_{P,tgt}}{N\sigma^i_t}\cdot CF(\bar{\rho}) $$

III. Backtesting Results

This section lists some interesting backtesting and analytical results of copper.

Overview of Copper's supply and demand

Overview of Copper's fundamental data

Results

• Inventory

  • Exchanges' Warehouse Inventory

    The warehouse inventory of SHFE reflects the consumption of copper better than that of LME, due to the matter of financing lock. The inventory of SHFE also shows evident seasonality.

  

  

  • Social Inventory

    The social inventory of Shanghai free-trade-zone is a good monitor to see the imbalance between supply and demand due to imports and exports, because Shanghai free-trade-zone is the buffer between Shanghai and London.

    When the losses of imports are huge, the copper accumulate in the warehouse.

    When the import window is open, the copper flows into mainland market from the free-trade-zone. For example, The inventory of copper fell 220,000 tones in the second quanter, from 400,000 tones to 180,000 tones.

  

  

• Price

  • Basis Rate = (Spot price - Future price) / Spot price

    

  • Refining charges

    Refining charges are the payments from mining merchants to the refineries. High refining charges means the abundant supply of the copper mine.

    Some factors of refing charges performs extraordinarily well in 2020. The copper miners went on strike for many times due to the pandemic. Hence the supply of copper mine was in extreme shortage and the refining charges was historically low in 2020. In this way, the subject of refining fees reflected copper's supply chain logic, which had an essential influence on the price of copper in 2020.

    

    

  • Import profit and loss

    The transaction premium of Yangshan copper reflects the premium to buy copper due to the supply and demand strucutre. Hence, high premium demonstrtes the situation that demand exceeds supply.

    In second quarter of 2020, the import window was open for most time. Strong import needs pulled up the premium and decrease the inventory in free-trade-zone rapidly. The inventory of copper fell 220,000 tones in the second quanter, from 400,000 tones to 180,000 tones.

    

IV. Return Models

Performance of Single Commodity

1. Copper

The correlation between the styles of factors is as follows:

	price	inventory
price	1
inventory	0.236	1

2. Aluminium

The correlation among the styles of factors is as follows:

	price	inventory	demand
price	1
inventory	0.254	1
demand	0.026	0.097	1

3. Zinc

The correlation among the styles of factors is as follows:

	price	inventory	demand
price	1
inventory	-0.036	1
demand	0.181	0.032	1

4. Nickle

The correlation between the styles of factors is as follows:

	price	inventory
price	1
inventory	0.010	1

V. Risk Models

Performance of Portfolio

• Performance of each commodity: before target volatility strategy

• Performance of each commodity: after target volatility strategy

• Performance of strategy

• Historical Performance

• Seasonality of the portfolio

VI. Future Work

1. Try other weight-tuning methods, such like spearman-IC value method of weighting.
2. Explore effective factors by cross-sectional test among all the commoodities.