I will continue this thread here.

to finish the continuous positioning simulation, I have to debug. on each given day, I need to find out:

> the target size
> the actual size
> adjustments to the size
> costs of the adjustment
> balance adjustment

this is boring but necessary

https://twitter.com/pedma7/status/1896127982206603611

lets look at the first day.

the weight of the signal tells me to allocate roughly 99% of size.
the realized volatility is around 68%.

so the target position size is (target_vol / max_pos) / current_vol * port_size * signal weight

(0.4 / 1) / 0.68 * 100 * 1 = $59

so the next task is how I turn this trend signal into a weight I can allocate to.

I remember this tweet from macrocephalopod where he mentions a sigmoid function.

it's pretty straight forward so let's implement it to the signal.

https://x.com/macrocephalopod/status/1587591587722559489

now once this mapping is done , I don't know when to change my allocations.

in a perfect world of no costs, we'd do this as much as we could, every second if possible.

but trading costs money so we need to trade when we have to.

https://x.com/macrocephalopod/status/1587896137918316546

so if we measure the continuous signal of a simple ma crossover (for simplicity) , I can see that it is pretty noisy.

there seems to be some relationship there, especially on the tails , but pretty weak close to 0, as expected. however there's something missing here.

I am measuring forward absolute returns, which is influenced by the asset's own volatility at the time.

earlier points will have higher vol as market caps were lower and things were a lot more jumpy.

3,923% compounded returns from this crypto strategy.

Not from gambling or luck, but from a simple systematic approach.

The best part? It only had a -27% max drawdown.

Here's the full breakdown of this strategy: 1/ Total returns comparison:

• Strategy return: +3,923%
• Benchmark return: +2,541%
• Market return: +1,610%

Strategy outperformed the market by 2.4x and benchmark by 1.5x over the testing period.

You are being lied to about your trading strategy's returns.

Traders waste 1000's of hours coming up with strategies to beat benchmarks like the S&P500.

But even when they find them, do they really beat it?

Here's what traders should be looking at instead: 🧵 1) What is Beta?

Let's start with something that a lot of people misunderstand.

If every time that SPY goes up by 1%, your strategy goes up by 2%, your strategy has a beta of 2 to SPY.

Simple enough, but let's go deeper.

This is Gary Stevenson.

In 2011, he was Citibank's most profitable trader in the whole world.

He also:
- Traded nearly $1T a day at his peak
- Became a millionaire at age 26
- Retired at age 27

His trading strategy? Borrowing and Lending Money:🧵 1) Performance

By the end of 2011, Gary was Citibank's most profitable trader in the whole world.

All from one single bet that society would collapse.

Despite this, Gary wanted to leave his position as a trader.

This is David Tepper.

In 2012 he received the largest single paycheck in the world for a hedge fund manager:

$2,200,000,000.

In the 1970's, he paid his college tuition trading an options arbitrage strategy, and today his fund manages over $6B.

His strategy? Buying bad debt: 1) Trading to Pay Tuition

In the 1970's, David was looking for ways to make money.

He noticed that option prices were often slow to adjust to changes in the underlying securities.

He saw this as an opportunity, and paid his entire school with this one strategy.

This is Jerry Parker

At 25 years old he was given $1,000,000 to trade.

He had never traded in his life.

But in the next 5 years, he and the remaining group, made over $175,000,000 trading.

9 lessons from 36 years of trading experience: 1) Losing 60% in 1 Day

Comparing returns on absolute terms to other traders is just an illusion.

You don't know what risks they are taking, you don't the how close they are to blow up.

Always keep risk in check and don't allow greed to destroy your portfolio.

If you think that using a fixed position size on every trade is optimal, you are wrong.

There's a better way to approach position sizing.

Here is a simple explanation of the Kelly Criterion and why it's so important to consider its benefits, but also its risks↓ 1) What is the Kelly Criterion

The Kelly Criterion is used to calculate the optimal position size for a given event.

For that we need 4 variables:

a) Equity Balance
b) Expected Return
c) Win Probability
d) Losing Probability

How do we put them together?

Everyone talks about edge decay.

There's a few metrics I like tracking, to ensure my models are performing adequately.

Sadly, as traders, we bet on the unknown, and to think we have control, is an illusion.

Here's a few metrics I look for in potential edge decay.. 🧵 1) Historical Limitations

Before we move into the stats, I want to make the point that historical analysis is in nature limited, and the future is unknown.

As traders, we are paid for the uncertainty of the unknown, and what happened in the past, is not definite.

Back in December-2023, I was holding the largest trade of my career.

More than 300% return in a single coin called TRB.

What allowed me to keep most of the returns?

Portfolio Rebalancing.

Why is it important and how to do it properly. Let's dive in: 1) What is Rebalancing

Let's imagine that we have $10,000 in our portfolio, across 10 different stocks.

To keep things simple, we assign an equal weight to each stock.

$1000 position per stock.

However stocks are volatile and some will experience strength while others lag.

QUICK INTRO TO SHARPE RATIO

This is a quick guide for anyone interested in Sharpe Ratio or already using it and trying to make sense of some of its nuances.

Bookmark it for your own study.🔖 1) What is the Sharpe Ratio?

In theory, the more risk you take on, the more you should be compensated for it.

It's not a good bet to take a bunch of risks, without the proper compensation.

That is what the Sharpe Ratio helps us understand and benchmark for.

Overfitting is a silent portfolio destroyer.

Your model looks great on paper, but not in real life.

Here are 5 simple ways to ensure your model is not overfitted: What is Overfitting?

In simple terms, overfitting means your model is too good at predicting returns in the past.

But not so good at predicting returns in the future.

It finds patterns in the noise, not the signal.

It took me 7 years to realize what it takes to become a great trader, and I will tell you in 2 minutes.

1. Even a bad or mediocre trader can make a profit over many years. That is not the definition of a great trader. 2. The definition of a great trader, is that who can deliver higher risk adjusted returns than a benchmark, on a consistent basis.

If you think that a backtest is enough, you are wrong.

Until you have done a proper audit, you don't know if your strategy's performance is truly accurate.

Here is my 7-step framework to ensure that my backtests are completely accurate before I deploy them↓ Step 1) Manual Check

The first check is the most basic one. But because it’s basic, it doesn’t mean it lacks importance.

Actually if you get this one wrong, the entire model is wrong.

We want to compare the trades our model simulates, with the actual trades that we’d have taken, given our parameters.

Do a minimum of 5-15 random tests on your trades. Ensure variability of the types of trades.

I'm a systematic trader.

For the past year I've researched 80+ trading strategies, for myself and my clients.

I found that most failed trading strategies all dance around the same problems.

Here's the top 5 mistakes when building trading models and how you can avoid them: 1) Complexity

Probably the most common mistake I find is this need for complexity.

Why is it a problem?

Most of them have never produced a single dollar from their trading.

But they're already working with machine learning, regime forecasting, complex math models, etc.

By 2021, Kristjan Kullamägi had made millions trading Episodic Pivots.

3 years later, he claims that the edge is still there.

Everyone seems to agree with that.

It's one of his most known strategies and everyone needs to know if it still works: 🧵 1) Kristjan's Story

This thread will be solely devoted to the study of the Episodic Pivot setup that Kristjan mentions on his videos.

If you don't know who Kristjan is, and want to know more about his story, the validation of his returns, etc, here's another thread I've wrote:

https://twitter.com/1256906641158680578/status/1775121586535678224

Covariance Matrix is often mentioned in quantitative trading research and papers.

But 99% of traders and investors don't understand it.

Here's how it works in a simplified format: 1) What is Variance?

Variance is defined as the average of the squared differences from the mean.

It basically tells us how spread out the numbers in our data are.

Now you may be wondering why do we square the differences?

Looking for new trading strategies?

Here are my top 10 places to look at in 2024: 1) SSRN

This is a website where you can find all research papers from all kinds of fields.

And yes, also the trading field.

You can find a lot of decent ideas here, even though you have to dig a lot to find gold.

7 in 10 traders blow up their portfolio.

That figure is probably even higher.

I had to find out their portfolio-destroying habits.

7 ways these traders think about risk so that you can do the opposite: 1) Sizing Too Much After Recent Wins

Newer traders get excited about profits, and scared with losses.

Recent trades lead them to oversize positions, without consideration for the full scope of risk.

In this example, the trader bet 40% of his account, in a single trade.

Linear regression is one of the best tools in quantitative trading research.

But 99% of traders and investors don't understand it. ❗️

Every trader needs to understand how it works: 💡 1) What is a Linear Regression?

Imagine that we want to predict a person's weight (y), using their height (x).

Linear regression is a statistical method to model relationships between variables.

The goal is to find a linear function that best predicts y from x.