After I had success betting on the oil price with a highly correlated investment fond, I came to the conclusion that negative correlations could be used to bet against the price of other assets. Unfortunately, it is not easy to find correlations between assets if you don't know which assets to compare in the first place.
So I created a website where you can find the 10 highest and 10 lowest correlations of certain assets.
But put options? Surely all you can do is lose what you bet in the first place?
Also, prices for these options (Tesla is a good example) aren't cheap. For example a bet that Tesla will be below $900 by Jan 19th 2024 will cost you about $250 / share at the moment. That means Tesla needs to actually be below $650 on Jan 19th 2024 before you make 1 dollar.
As long as you understand whats going on and what you are actually betting on, you are fine. But options have lots of 'gotchas'.
https://en.wikipedia.org/wiki/Put–call_parity
i.e. When you short a vanilla equity, you'll likely have transaction cost of borrowing the stock as an interest rate over time. This cost is incurred as you keep the position open. This cost is related to the cost of capital for the shares you've borrowed. Cost of capital is an implicit cost on time.
To construct a synthetic short (using options/bonds), you basically short a synthetic equity. A synthetic equity can be constructed through a long call position, short put position, and long bond position. This synthetic will mostly replicate the the stock's return. To turn it into a synthetic short, you just do the reverse, short a call, long a put, and short a bond.
A bit like Negative Chess...
But unlike selling stuff you don't have (shares, options etc), you won't lose more than your stake. You have $1k, you put it on red, you invest in the S&P, you buy Roubles, you buy a meme stock, you buy a TSLA call for 6 months time at $2k, you buy dogecoin, at most you will lose $1k.
You borrow 100 shares of TSLA when it's $900 and sell them, you'll get $90k now, but you might end up with TSLA jumping to $5k overnight and you have to send those shares back, that will cost you $500k, well done.
The bigger risk comes in selling options: you sell a call and (assuming you don't already have the underlying shares) your downside is unlimited - you might have to pay sky high market prices only to sell (to the call option holder) low; you sell a put and your downside could be the entire strike price - the underlying could have gone to zero but you're forced to buy high (at the agreed price).
I got myself mixed up and was misled by a Corporate Finance Institute quote (which incorrectly notes the loss is unlimited only to contradict itself):
For the seller of a put option, things are reversed. Their potential profit is limited to the premium received for writing the put. Their potential loss is unlimited – equal to the amount by which the market price is below the option strike price, times the number of options sold.
The value of a call option is unbounded.
This link[1] has some good questions and answers about options, although it certainly isn’t complete (e.g. dividends are mentioned but voting rights are not).
[1] https://www.blackwellpublishing.com/content/kolb5thedition/c...
[2] https://news.bloombergtax.com/tax-insights-and-commentary/ma...
- Building things like this is always great. And its a fun site to poke around on.
- I would not count on this approach or expect it to be reliable in terms of actually hedging. Correlation, as a measure, has lots of issues. You are boiling down a lot of complex relationships into a single number. While it is convenient for many calculations, there are many problems. For example, many asset classes will go through periods with positive correlation and then later, negative correlation. This is due to a factor driving both securities price becoming more or less volatile compared to the other drivers. E.g., recent increased volatility around inflation expectations driving correlations between rates and equities. Whereas, few years ago, inflation was not driving anything.
- One alterative approach is to have a "risk model". Which essentially decomposes a security into drivers. Each security then represents a basket of these drivers. You can then use this model for range of purposes. While not perfect by any means, the model contains more information than a correlation. These too have a range of issues and creation and use is as much art as science.
- In general, you won't find many negative (or even very low) correlations across individual equities. Most stocks are driven by a common set shared risk factors that drive much of the risk. But if you can find negatively correlated securities (or lowly correlated), then that is certainly helpful.
Is it more likely to find negatively correlated securities across industries, like the SPDR ETFs, rather than individual stocks?
And most people blow up their account, because their trading strategies worked for a long time.
Speaking for a friend, of course.
Having said that, if you can find a negative correlation AND there is a strong reason to believe that the negative correlation will hold as long as X and Y hold, then the tool could be very useful.
The TLDR is more that correlations throw a lot of information away.
However, I get a “504 Gateway Time-out” error for https://betagainst.fun/asset/bz__f_bno/. HN hug of death?
- The acquiring company's share price drops because it often pays a premium for the target company, or incurs debt to finance the acquisition.
- The target company's short-term share price tends to rise because the shareholders only agree to the deal if the purchase price exceeds their company's current value.
As an aside, the methodology says "We calculate the correlations between 2 securities on the daily closing values of the last 20 years. If one of the two securities has not been on the market for so long, we use all available prices to calculate the correlation." - hopefully the author means daily changes, not daily values, because otherwise everything is spurious.
That said, in any set of 20+ variables there will be a 10 highest/10 lowest correlating.
Without a good (specific, hard to vary) explanation as to why the correlation happens, I would not use this information to gamble real money.
If you track this over time, you will see that these change (often at times when you really want them not to).
Past observances aren't super helpful in predicting fractal futures.
[1] https://papers.ssrn.com/sol3/papers.cfm?abstract_id=525282
The idea is not to have perfect negative correlations, but to find values with some negative correlation. This way you can profit from falling prices of an asset. Not with a 1 to 1 yield, or even higher.
If your goal is to manage the maximum loss from a short position, it’s much easier to short the stock and buy a protective put. This gives you perfect exposure to your bet, while allowing you to manage your loss exactly.
You cannot compare the same or roughly the same basket of assets against itself.
It's better to correlate daily returns than daily prices, since the latter are nonstationary, and I suggest using 1 year of daily returns rather than 20 since correlations do change over time. When I worked as a financial quant no one looked at 20 year correlations to measure near-term risk.
As such the holding period might be on the order of 1 year for some people, so just 1 year of daily returns might invite too much trading if you're rebalancing.
A little tip from someone who dabbles in algorithmic trading, look into cointegration as well as correlation. Also, the cross correlation matrix changes over time, you can have great fun seeing spikes and convergence/divergence as markets tend to get more or less correlated reacting to real life events.
With few exceptions, all stocks are correlated to the broad market.
https://en.wikipedia.org/wiki/Beta_(finance)
One explanation that would be popular right now is that the price of stocks can be calculated based on expected future cash flows and interest rates (e.g. think of what you'd pay for a bond that produces the same csh flow.) Since the interest rate is a factor in that calculation for all stocks, there's a correlation right there.
The exceptions tend to be sketchy. At one time it was thought that gold mining stocks would move against the market, but what I heard was that gold mining firms are badly run and if you like gold you should just buy gold.
To me I don't care if a certain stock is correlated by something, I would more like to know which stocks do have correlations or if there are correlations with a time lag
Of course, the problem with something like this is that using it would, of course, ultimately change the market so that x→0.
Prices are not adjusted, as these are not free.
The searching UI for companies has no loading or success/failure indicators. A note on performance - waiting an extra second before firing the search request can help take some of the load off the server, along with cancelling requests after I change the search.
Some requests time out, and others return no results (e.g. HOOD, TLRY).
For other requests that returned results, there's no correlation (Apple).
Maybe an example page could be helpful to illustrate what the app can do?
Normally, the search works instantly. Even though, I am located in Germany and the server is located in the US.
I debounce the search for 250ms.
Mispelling “higest” in tab.
A convenient tool. I’m restricted from trading in my employer’s stock. There are no rules about trading in a highly correlated proxy.
Also, typo: Higest -> Highest
Where are you getting the historical data? There's a ton of fun stuff you can do with this kind of long-horizon of data
These are starting to look like perpetual motion submissions to the patent office.
I am supervising a master thesis project (in fact the second such project on the matter) where we are trying to predict the covariance matrix of a portfolio of assets using machine learning. Results are promising!
https://factor.fyi/questions/top-10-aapl-correlating-stocks-...
Little offtopic highjacking: does it make sense to "bootstrap" correlation among stock returns (frankly, any multidimensional time series, but since we're talking about stocks) with different time periods?
Say, for any pair of stock a and b, randomly selecting a startint point and a period (N days) and using this as a better estimator for the "true" correlation instead of using all the data points? Or something like this, not this process exactly
Cool idea, by the way. I'll have to play with this more.
One small typo: Highest -> Higest on the graphs
I thought the opposite of buying a single stock (Alpha) would be in service of.... Beta Gains T.
