Fosse possibile...

stai dicendo che non è possibile oggi?

Here are my issues with this strategy:

1. This strategy doesn't account for Volatility Drag (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1664823). The largest obstacle is to overcome periods when implied vol is on the rise while realized vol remains relatively calm. As this article's strategy employs a spread, it will get eaten alive twice as quick by geometric compounding...thus the article's "Conclusion" is wrong.

2. Your time frame for testing is too small. What would happen in 2008? What would happen in 1999-2002? What would happen in 1987? What would happen between 1974-1983 (sideways)? What would happen in 1929? And so on.

3. In a large magnitude tail event, this strategy would wipe you out if incorrectly positioned.

4. The article's optimized parameters scream of over-fitting.

I employ a similar strategy but account for the above pitfalls. This article has a distinct lack of understanding of both volatility and mean reversion.

The results speak for themselves. Your comment shows that you lag the basic understanding of how these ETFs operate. What you call pitfalls is exactly what is giving you the opportunity to trade. Since when are short term strategies tested on decades of historic data ? Ever heard of market cycles and are you aware of the fact that for your mentioned time period these ETFs didn't even exist.

Francesco Placci

I say you cannot conclude anything with only 62 trades.

Significance tests are performed to see if the null hypothesis can be rejected. If the null hypothesis is rejected, then the effect found in a sample is said to be statistically significant. #Statistics #MonteCarloSimulation

I know how to exploit volatility just fine... It's called bet against Mad Mr Market. Sell insurance on events that are more than 2 standard deviations outside the norm... My tenets are a) the market is inherently under damped (as a control system) and therefore oscillatory. Ie it can't help but go too far. And b) the market is mean reverting. So bet on long term sanity and bet against short term insanity. In practice put spreads and call spreads are my means of doing this but you need to optimize on a) the probability of exercise vs the value of the premium collected, b) the margin that is required, c) correlation between various positions, d) stable versus unstable stocks which relates to item a).

So long term mean reverting and short term volatility. In the world of options an interesting approach as long as you account for variable change. Karen implemented such a system with great success. Have a look at the interview https://m.youtube.com/watch?v=BquDGE9KxZQ

The economic reasoning behind the strategy is valid, short vol when in contango long vol in backwardation. However your implementation is missing critical information.

1. What is the time period for the backtest?

2. Did you optimize the strategy on the same time period you backtested?

3. Where are the out of sample results?

4. 100 monte carlo runs is not conclusive, what about 1000 or 10000?

5. Is this strategy tested on fictional consolidated close prices or did you get real quotes?

6. Did you cross validate your optimized parameters?

7. How stable are your strategy results if any of the parameters are changed?

8. Have you considered transaction costs? Shorting a derivative based ETF can be expensive.

These are just some of the questions I had when reading your article.

1. 4 year time period.

2. Yes.

3. Out of sample tests where done.

4. 1000 simulations on 62 trade signals ? Enlighten me.

5. Last price traded.

6. Cross-validation gives a measure of out-of-sample accuracy by averaging over several random partitions of the data into training and test samples. Yes=3

7. Look back period is very stable up to n=35 after that the results are not consistent.

8. Yes of course.

Marko, I think you need some respect for putting your strategy public even if the questions of Brian are legit and statistical significance on 62 trades is a question mark. About Karen, be careful with that : http://macro-ops.com/karen-the-supertrader-goes-rogue/

She didnt account for the Short Vomma Position she was taking which mean whenever the IV was rising, which cause you Short Vega Risk to rise dramatically.

If the strategy sounds logical then why are 62 signals representing an issue. If this is true then we should abandon the whole theory of seasonal trading patterns.

You mean she didn´t account for variable change. Accounting for this too is like creating the "perfect" system. Thank you for the input will be part of the next research topic.

I just recreated this system and had a very hard time until I discovered you bound the spread then it made some sense. Your average days in a trade is 1.05? typically in backtesting if you are exiting the same day as you enter will give erroneous results because of bouncing ticks. The profit factor looks good and is not overly optimistic.

Marko, do you currently trade this strategy live?

No, i am still working on this strategy. Fine tuning is required.

The glitch is the execution price should be buying at offered and selling at the bid.

When you use these prices, profitability is affected.

As for trading volatility. There are many measures. You can have a moving weighted average of the last 100 trades, 50 trades, or less. Then you need higher moments, what is the change in volatility in the different time lags. Or you can calculate volatility on "market" executions. Or only on a volume trigger . So many ways.

Most importantly, a model may have a prediction at all points, and continuously. That is not useful. A signal may be given by one of your 50 models, say, once a day. Not continuously. Incorporate and identify events. And your Bayesian probabilities could be tampered by an event trigger. Think about having a firm opinion for every topic in a conversation. Silence is golden, especially when you should not trade. Thanks for your comments, you folks are golden.

Backtesting is a reasonable way to confirm your equation , at the end of the day there is no way around real Trading, only the market will proof your model .....

Have no fears, whatever strategy used in trading is superb as long as it produces satisfactory results. Keep working on the strategy till it can go live - that's the final test in defending your strategy. Well done Marko.

Ralf : back-testing is a method by which you explain the past. You can cast your results to the future by conditional Kalman filtering or Bayesian mapping . I like to use a fox chasing a rabbit. The fox tries to build a model of the rabbit's evasion tactics. Based upon passed behavior. Market prices, also adjust through arbitrage and price competition, and are chased by players who attempt to forecast the future. What is the flaw ? the rabbit also learns. So each clever trick or "neural net dynamic adjustment forecasting artificial intelligence high frequency quantum filtration forecasting profit generating algorithm" assumes that the potion is unique and others are unaware and will not adjust. Hint: what is the solution to this conjuring trick ? Are you a fox , a rabbit, a.... ?