Prev Article: 1.1 - Quantitative Strategy Types

Searching for Strategies

How do you find good strategies? Here we will look at some considerations:

We will look at two different approaches in finding strategies. Data Mining and Idea-Based.
We will look at the edge for strategies to be profitable.
We will look at different trading styles and how it affects how you choose strategies.

Data Mining vs Idea-Based

There are two main different camps of building strategies: Data Mining, or Idea-Based.

The data mining camp goes like this: "I wonder what factors drive price movements, so let's sift through the data to find these factors."

Example: Testing the correlations of price data to each of the hundreds of macroeconomic factors.
"I want to look for stories from the data that I have"

The Idea-based camp goes like this: "I have a hypothesis about how the price moves, and I want to test and design a strategy based on this hypothesis."

Example: Testing the correlation of price data on GDP growth, because you have a suspicion that it is correlated.
"I have a story, but does the data say that it is true?"

Different traders prefer different approaches. That said, there are glaring problems with the Data Mining approach:

Multiple Comparisons Bias -> By running too much experiments, you run the risk of getting a random strategy that is statistically significant, despite being random.
- For example, wikipedia page of Multiple Comparisons Bias shows the high correlation between "# People Killed by Venomous Spiders" and "# Letters in winning word of National Spelling Bee". Spurious correlation, to say the least.
Overfitting -> By running too much experiments, if you are not careful to not use different test sets in each experiment, you run the risk of "implicit overfitting". Dangerous because especially in finance, test data is not always representative of real data (which is ever changing).
- If we run 100s of experiments on the same test data, and then picking the best result from those experiments, isn't it just the same as fitting to that testing data?
  Some traders advocate Data Mining, but it requires deep handling of the above problems (often with more advanced methods).

The idea-based approach is also not free from problems if you use it wrongly.

There is the problem of P-Hacking, which is basically the same problem as Overfitting above. That is, you run multiple experiments, and select the one with the best P-Values.

Profitability

In choosing strategies, you need to understand the profitability characteristics of that strategy. Why is it profitable compared to other strategies? Are there any risks that you take by choosing a more profitable strategy?

Before diving into what makes different strategies profitable, below we explain the main reasons why strategies are not profitable

Overfitting -> Turns out your strategy only works on the test data, which is different from the real data. Tough Luck.
Regime Change -> Turns out the data you used to design the strategy was out of date, and the market has shifted regimes (e.g. from growth to value, etc.). Markets change constantly, and the reality is you need to recycle your strategies to keep your trading profitable. Tough Luck.
Alpha Decay -> As time goes by and traders experiment with tons of strategies, atraders will converge on the same profitable strategies. But by having more traders using said profitable strategies, the amount one can gain from the strategy will reduce. Tough Luck.

As for what makes trading strategies profitable:

Risk Premium -> By taking on certain risks, you stand to gain more. For example, if you buy during blood on the streets panic of 2009, you would make a huge amount of money when the market goes back the subsequent years. But are you willing to take the risk?
Leverage -> Having leverage means standing more to gain - and lose. It's like intentionally injecting risk premium into your trades.
Liquidity Premium -> Investors happy to hold less liquid assets (e.g. land) will get greater opportunity for higher returns. But ivnestors in very liquid assets (e.g. money markets) dont get the opportunity for high returns.
Hard Barrier to Entry -> The harder to get into, the higher potential returns. For example, a HFT trading in the millisecond timeframe will be much harder to get into that the weekly timeframe trading, and thus less crowded.
Behavior -> People exhibit behavioral patterns that we can exploit. For example, people tend to buy during price increases, and sell during price decreases.
Actual Alpha -> Perhaps after much research and testing you've found some strategy that successfully outperforms the market. That's great!

Regarding Risk premium, there is also the concept of skew:

Negative Skew: Consistently positive returns, with occasional high drops.
- Example: Selling insurance. Frequent positive returns, but occasional big payouts to clients.
- Equities are mildly negative skew
Positive Skew: Balanced positive and negative returns, with occasional high increases.
- Example: Buying insurance. Frequent small losses from buying insurance, but occasional large gains in times of emergency
- Gold is positive skew.

Trading Styles

Knowing the profitability of your candidate strategy, the next step is to choose one that suits you. To do this, you need to understand your own trading style.

Below are a few trader archetypes.

Static vs Dynamic

Static -> Buy and hold. Risk characteristics are directly inherited from underlying assets.
- Side Note: Other types of static investing is "Rebalance Investing" (where rather than buy and hold, you rebalance to set a constant allocation) and "Risk Parity Investing" (where you rebalance to set equal risk between your assets)
Dynamic -> Buy and Sell according to your rules. Risk and return characteristic also depends on your strategy.

Positive vs Negative Skew Preference

Positive Skew -> Frequent small losses and infrequent large gains. Because losses are frequent and small, risk management is more predictable.
Negative Skew -> Frequent small gains and infrequent large losses. Because losses are large, risk management less predictable.

Fast vs Slow

The speed of your trading is determined by your average holding period. How long do you prefer to hold before you trade on your positions?
Most trading rules have decreased Sharpe Ratio if their holding periods exceed several months.

Technical vs Fundamental

Technical -> Relies purely on price data.
Fundamental -> Also relies on micro and macro data. Barrier to entry is higher because fundamental data might not be easy to access and parse.

Portofolio Sizes

Large portofolio holders are exposed to liquidity risk, because their position is big enough to affect the market price. Thus they need to spread their trade over bigger asset classes.
Small portofolio holders need not worry about this too much.

Contrarian vs Follower

Contrarians -> They take advantage of mispricing - i.e. they believe that the price will eventually go back to some mean. Tends to be more negative skewed. Examples would be mean reversion strategies.
Follower -> They take advantage of the current market movement - i.e. they believe that the price will continue to follow the direction of the market. Tends to be more positive skewed. Examples would be momentum strategies.
For more detail, see the section #Strategy Types

Instruments

Choosing a strategy is good and all, but we need to implement the strategy on some instrument class.

Here are a few considerations when choosing instruments.

Data Availability -> Can you get access to the data?
Do you understand the instrument?
Volatility -> A more volatile asset means a potential for greater returns (because the standard deviation is higher), but also means you need a stronger risk management.
Correlation -> Highly correlated asset returns are dangerous, because both asset tends to move together. When both are affected by the same downturn, your portofolio moves more wildly. Diversify, choose assets with uncorrelated returns.
Cost -> Trading costs can take up the bulk of your profits if you are not careful
Liquidity -> Mostly important if your capital is big. But low liquidity also means more volatile price changes.
Returns Skew -> Negatively skewed assets need more careful handling, because of its less predictable downturn nature.

Sources

Next Article: 1.3 - Backtesting