Hi folks,
This question also puzzled me for a long time. I would like to share my understanding on this issue. I think some previous posts have already shed some light on this. Hopefully, my answer is more systematic and can make things more clear. Of course, any criticisms and suggestions are welcome.

My short answer is that there is no “absolutely correct” way to handle reinvestment ratio in order to fully maximize your gain without getting a margin call. The SR rule is more a subjective tradeoff choice rather than an objective rule.

First of all, SR rule is derived from a diffusion model. No model can really model financial markets or how a trading strategy exactly progresses in a financial market. SR rule is itself just one way to protect your account.

Secondly, we take a step back. Let’s just assume this diffusion model is a pretty accurate approximation on how a trading strategy will behave through time. Usually, we use max drawdown or maybe CBI index as a criterion to stop a trading strategy when experiencing drawdowns. If you use a higher reinvestment ratio (reinvest all profit generated so far, and thus higher trading lots), it is more likely your strategy will be expired given a usual expiring criterion (20% percent drawdown or 5% CBI index). If you use a lower reinvestment ratio and thus lower trading lots, you essentially give the strategy more chance and time for it to survive through current adverse market condition and work its way back into making profit again. So, the reinvestment ratio is the tradeoff between making more money when the market fits your strategy and the survivability when the market hits you hard given some drawdown related strategy expiring criterion.

Thirdly, if we think about the strategy development process, more things are subjective rather than objective. In strategy development process, whatever your approach is, backtesting is a step you can not omit. But the question is how long the history data you should use to backtest. Let’s say you want to develop a strategy that can not have more than 20% drawdown and you use 5 years of history. Once you have successfully developed such a strategy, you now set the trading size to the point where 20% drawdown is the maximum during the backtest. However, what if you test 1 more year back in history and you get a 30% drawdown, now should you use the previous initial trading size or modify the trading size corresponding to this new backtest result? If you do not get such drawdown in 1 more year, what about 2 more years back, or 5 more years back? Theoretically, your drawdown will increase given a fixed trading size when you backtest longer history, and thus your starting trading size will decrease when you go live. So, my point is that even the initial trading size (or required capital) is in fact subjective. Therefore, the SR rule based on the initial trading size will also be subjective.

Hopefully I am making some sense here. To answer the initial question, should one use SR rule after stopping a strategy and resuming it, or stopping and waiting some period of time then resuming it, it really depends on your knowledge and confidence on the particular strategy. However, if you don’t have enough information or confidence about a strategy, it is probably better to use SR rule to be safe if you know how much profit one strategy has already made. If you don’t know such information, you better test longer history to be on the safe side. After all, based on years of JCL’s experience on such matter, I assume his advice would be very valuable.

Valid arguments. The SR rule assumes a random distribution of wins and losses. This is normally a good model, but only a model, and real systems can deviate from it.

Even when the model is correct, we must deal with two facts that appear contradictory to the human mind. The longer you trade, the higher the probability for you to have encountered a bad drawdown at some point. But at any given day the chance of a bad drawdown is exactly the same as the day before. So the day we start a system has no special meaning. The bad drawdown can happen within the first week of trading with the same probability as in a week after 10 years trading the system. It only matters how long you traded, not when you started. Thats why the SR rule applies to trade profits, because for having them you must be already trading for some time during which you're exposed to the drawdown risk. In some situations the SR rule is not obvious:

You got two systems A and B. You invest $1000 in A. After a year, the account is at $2000. Now you stop A, go on a long round-the-world trip, then come back and start B. B has never been traded before. Do you invest the full $2000, or only $1440?

Your aunt has traded system A and has doubled her account to $2000 after a year. Now she gives you the $2000 and suggests that you use it for trading the same system. How much do you invest?

You know that your aunt has traded system A and has doubled her account to $2000 within a year. By chance you also have $2000 on your savings account. You now want to use it and start trading the same system. You have never traded before. Do you invest your $2000, or only $1440?

I know this is an old thread but I would like to ask an additional question to your scenarios, in particular this one:
You got two systems A and B. You invest $1000 in A. After a year, the account is at $2000. Now you stop A, go on a long round-the-world trip, then come back and start B. B has never been traded before. Do you invest the full $2000, or only $1440?

Let's say I trade one of the Z-systems and then with a new release of Zorro, I update the .par and the .fac files. Am I now trading a "new" system? Is this a similar consideration like the above example?

Thank you!

It is a similar consideration and it is not relevant whether it's a new or old system.

Thanks.

With regards to the whole model itself, the blog post https://financial-hacker.com/build-better-strategies-part-3-the-development-process/ it says that "On leveraged accounts with no limit to drawdowns, it can be shown from statistical considerations that the maximum drawdown depth Dmax grows proportional to the square root of time t".

--> Is there a source you can link to where this model/this statistical consideration is explained?