The reasoning behind "rescaling" strategies

C2
1

I’d like to correct a few misconceptions that have been stated here, again and again, regarding the “rescaling” of strategies.

First, what is strategy “rescaling?”

A strategy developer “rescales” when the nominal size of his C2 Model Account becomes too large for him to manage, or for subscribers to AutoTrade.

Here is a typical example. A strategy starts off on C2, and it trades 1 or 2 or 3 futures contracts per trade. It starts with $50,000 as its nominal Model Account size. Happily, after a year of trading, due to successful trading, it now has $150,000 in its Model Account.

The first thing to notice is that this large capital basis is disconcerting to potential new subscribers. They might feel comfortable allocating $50,000 of their real trading account to a single strategy… but $150,00? That’s too large.

So that makes some potential subscribers shy away.

Now, let’s look at it from the perspective of the strategy creator. He has a strategy that has done well trading 1, 2, or 3 contracts. That’s his strategy. He’d like to continue trading in the way he always has: selecting 1, 2, or 3 contracts per position. But now that he has $150,000 in capital, if he trades that way, his potential return percentages go down. A one-unit trade on $50,000 of nominal capital might lead to a decent risk-adjusted return; a one-unit trade on $150,000 of capital barely nudges the needle.

Of course, you might ask, now that the strategy has $150,000 as his Model Account - an effective tripling of its original account basis - why not just triple the strategy’s trade sizes: i.e. why not trade 3,6,9 contracts?

Well, because a lot of subscribers don’t want to make trades that large. And while it is true that they can use a combination of AutoTrade “scaling” and AutoTrade max-size limits to make their own trade sizes smaller, under some circumstances, this can lead to tracking error with the Model Account due to rounding issues.

This is why Strategy Rescaling exists. The idea is to “give back” some of the extra cash in the Model Account that has been earned through trading, so that a strategy developer can trade his strategy using the trade unit-sizes he likes, and while simultaneously maximizing the capital efficiency of his strategy.

In other words, if a strategy likes trading 1,2,3 contracts, and it likes using 20% of its available Model Account capital per trade, then “rescaling” allows strategies to do this, even after the Model Account has grown from its starting size.

The best way to think about rescaling is to consider stock splits. For example, you may know that Netflix (NFLX) recently carried out a 7-for-1 stock split. This means it currently trades in the $100’s. But a few months ago, it traded in the $700’s.

Now, if you look at the NFLX chart, you’ll see that between February and April of 2014, the stock went from 63 to 48, a loss of $15.

But wait…

Back in February 2014, the stock was actually trading at $441. And it actually went down to $336. That was a decrease per-share of $105 … not $15!

Wait! Is Netflix being “deceptive?” Are they trying to “hide” the drawdown that the stock underwent during that period?

Of course not. Because everything is still the same on a percentage basis. If you were holding some portion of your net worth in Netflix stock in February 2014, that portion decreased in value by 23%. It doesn’t matter what the dollars per share number is, because the number of shares you owned changed post-split.

This is similar to what is happening here when strategies “rescale.” A historical max-drawdown of 15% remains exactly that… a 15% historical drawdown. The nominal scaling of the capital basis is immaterial.

Since all meaningful statistics about a trading strategy are percentage-based, the fact that a strategy rescales in order to keep its nominal size reasonable does not change anything at all. No one is hiding anything, and no one is being deceptive.

Now, in general, I don’t encourage frequent rescaling of strategies, because it is annoying to subscribers, who have to re-set-up their AutoTrading each time it is done. For example, doing it once a month is foolish. But sometimes it is a good idea, particularly after a long and large run-up in Model Account equity. I want strategy managers to be able to choose this path without receiving criticisms based on what appear to be misconceptions about the process.

Matthew

4 Likes
2

Mathew,
The problem is not with stock or forex systems so your examples of Netflix or stock splits is not relevant. The deception is only with Futures systems because they are almost always traded with 1 lot trade sizes. Take a look at this simple hypothetical stock vs futures system which trade the SP 500. I think it will explain the problem I’m talking about. If subscribers knew the actual current risk instead of seeing only an artificially reduced risk metric (max $DD based on fractional futures) they almost certainly would not subscribe to those systems (such as Maximus Decimus):

3

Matthew,

It seems to me that once a futures systems scales down the draw down in dollar terms as shown on the system page on the horizontal preliminary statistics line close to the top cannot be the real drawdown experienced before the system scaled down.

I don’t think you can use the analogy of a stock trading system. BTW, it seems to me that for a stock trading system it is totally unnecessary to scale down, it is much easier for the subscriber to scale to his capital base or comfort level and if you look at the stock trading systems it is done by auto subscribers all the time.

Karl

4

Matthew and AshM,

I believe both of you have made valid points. How about considering a new feature on C2 where re-scaling is indicated on the equity chart with a marker (similar to charts at Google Finance representing a stock split) or listed below the chart in a table?

AshM, do you think this would provide a better representation to subscribers?

Regards,
ACA

5

Can someone show me a futures system too large to the point it needs to get rescaled down ?

6

Perhaps adding a “split” indicator would help but you’d have to know the original account size so you could calculate the actual dollar drawdown from the percentage displayed. I think a better idea would be to only allow rescaling of futures systems if every trade was divisible by a whole number, that way everything, including the current lot size, can be scaled, not just the drawdowns in the past.

7

Hi Mathew,

I understand your point of view on this subject but I have an idea that while I believe will be quite easy to implement it will solve also the deceptive thing connected with futures: Just add in the main panel what have been the biggest dd in dollar term per contract suffered since the start of the program.
In this way you will address who try to show unrealistic past result and thus who try to get from this trick an unfair advantage.

An example should clarify the problem : I have a program with 100.000 $ and I trade 4 contracts. I loose 20% so 5000$ per contract and a total of 20.000$ per position

Then I rescale to 10.000 to trade my program with 1 contract . How is implemented now the rescale thing for futures, It will show a % dd of 20% that investor will decipher as a loss of just 2000$ (the capital currently show after the modification)when in reality we know is 5000$. Or in other term is 50% of 10.000 the capital allocated, PER CONTRACT.

8

What about the profits ? You could say the same about the profits no difference . Example : I made 20K with my 100K account trading 4 contracts i now scale it down 1/10 to trade 1 contract as per your example so it will now show that i’ve made 2K when in reality it was 5K/contract = 50% of - newly - allocated capital .

9

Although I do not think is a meaningful stats like the Maxdd per contract, a MAXprofit per contract can be implemented as well. The whole point I believe is to give a stat of the key figures on a contract basis for futures. Otherwise is just fantasyland if rescaling is performed.

10

I dont see the difference , you cant put a note on the max dd - 5K loss/contract - and leave annual and cumulative return at 20% = 2K when infact it was 5K/contract as well . If you do that it will give a misleading presentation of the system .

The real solution for rescaling is to not allow it if it leads to a trading history of partial futures lot .

11

I agree with you 100% on that. It appears there are 2 way:

  1. Avoid rescaling if it leads to a trading history of partial futures lot as suggested.

  2. If point 1 is not possible then the % must reflect the minimum tradable lot. In that case 50% and 50%. (previous example)

Make sense?

12

So basically what we are saying here is when someone rescales his system down the scaling stops at the minimum tradeable amount = 1 contract and C2 will recalculate his statistics and results accordingly . I dont think MK or any developer will go for that :slight_smile: .

14

My two cents as a new user and somewhat experienced quant from a big hedge fund:

For me it would make more sense to report a cumulative return index instead of strategy NAV (dollar size). So, everyday you would start with the same fixed account size, you would base your trade size with this size, you compute return at the end of day based on this account size, add this return to the strategy cumulative return index ( Index(t+1) = Index(t) + return) ) and then for the next day the account size is again the starting account size (so basically rescaling is done every day).

This would produce a PnL that can be replicated and won’t suffer from fractional contract size fallacies. In order to handle different sizes of subscriber’s accounts, they would just rescale accordingly as their allocated capital to the strategy grows.

Instead of evaluating strategies comparing the stats on the dollar NAV, we could use this cumulative return index. Of course, C2 can always show simulations of how a given initial dollar allocation to the strategy would have grown in the past.

J

2 Likes
15

That sounds like a very interesting concept - I wonder how many people would go for it.
Matthew, any comments?

16

What will be the math if position is held for several days/weeks?

17

What about profits and losses. Say your initial account size is $50K. What do you do with the accumulated profits in real trading? More important, how do you make up for the losses to sustain your original unit size?

The answer is to create a VADI (Value Added Daily Index.) Multiply the index times the daily percentage change. The index records the compounded performance, and the developer can change the size or fee at will and not affect the index. This is common practice with CTA reporting, except they report VAMI (Value Added Monthly Index.)

1 Like
18

The VAMI used for CTA performance usually starts at 1000 and scales up or down from there based on compounded monthly returns. While we are at it, C2 should show this graphically as semi-log or ratio scale. It would also be nice if the little performance charts were all scaled the same. Then a 100% gain would be more obvious than a 10% gain. Now, everything is scaled to fill the chart box, and comparisons are misleading.

I know Matthew has a lot of projects ahead of this, but creating the VADI (Value Added Daily Index) would solve the problems with re-scaling.

1 Like
19

VADI could be a good complementary stat. The problem with using VADI here at C2 is that it can also mask trading fractional number of futures contracts.

The cumulative return index I propose is simply the cumulative $ PnL of the strategy. The PnL over several days is added. Every day, the position size and margin requirements are based on the declared nominal starting equity size of the strategy, which is static, it does not grow or decrease with the strategy PnL.

I think is better to keep a $ PnL, so that it is possible to quickly compare volatility or drawdowns to nominal starting equity. A strategy is too risky if its typical (or maximal) drawdown size is big enough to reduce the nominal starting equity to levels bellow the required margin for the strategy to work.

20

A water-tight and workable proposal that address the futures issue. Let’s see if C2 willing to comment about.

1 Like
21

The cumulative return index is not workable. Let’s say that an account is down 50%. The percentage gains, losses, drawdowns, and margin requirements would actually be twice the amount calculated on a nominal starting equity.

An index is the historical record of performance. Do Dow-Jones or S&P 500 go back and recalculate their index every time a stock splits or is added or dropped? Of course not, the history is locked in. The fractional contracts shown in C2 history tables would be of value to just show the source of the gains and losses. They were one contract or more when the trades occurred.

Using a VADI (Value Added Daily Index) starting at 1000 would make comparisons between systems starting with $10K, $100K, and $1000K viable. Rescale the account down (or up) or change fees, it doesn’t affect the historical VADI index. That is locked in. Don’t try to rewrite history as is currently done, which leads to all sorts of misleading distortions.

1 Like