vpFREE2 Forums

Bankroll requirements (was Max ER VS "other" strategies)

Steve Jacobs wrote:

I don't believe it is generally correct to change playing strategy based
on the size of your bankroll. One playes max-ER because the objective
calls for it, not because the bankroll size "calls for" or "allows"
max-ER strategy.

"Correct" in pursuit of what goal? If you are playing a positive game, and
your goal is to maximize how much you expect to gain over any fixed number
of plays, then a max-ER strategy will achieve that (unless the bankroll is
very low relative to the advantage and variance). If your goal is to
minimize the chance you will eventually go broke, the best strategy is to
not play.

There are a number of goals that are equivalent to one another, however,
that call for slightly different strategies depending on the bankroll.
Those are the Kelly criterion goals, and they include minimizing the
expected time to reach a particular level of winnings, or maximizing the
chance that you end up with more winings than any other prestated strategy
over a sufficiently large number of trials.

For these strategies, as the bankroll falls, the strategy becomes
increasingly risk-averse, as long as it retains positive expected
log-bankroll change. If the bankroll falls enough, that optimal strategy
will be not to play. If the bankroll grows enough, that optimal strategy
will be the max-ER strategy. In between, the strategy you refer to as
min-risk will be optimal for some range of bankroll, but except in a few
degenerate cases, it won't be the "best" strategy independent of bankroll.

One plays min-risk because they want the highest probability of having
their bankroll last until a goal is reached. Min-risk does this best
whether the bankroll is a single unit or a million units. This remains
true whether the goal is "play indefinitely" as per risk of ruin, or
the goal is "multiply initial bankroll by N' or just "build bankroll up
to G units."

Well, if the time allowed is finite, these latter goals are
Kelly-equivalents; a bankroll-sensitive strategy will dominate the
bankroll-insensitive min-risk strategy.

3) If the min-risk strategy says that the required bankroll is N units, no
other strategy can give a smaller required bankroll for the same goal.

No other fixed strategy can, but a strategy that adapts to the current
bankroll level may.

max-ER doesn't belong on an altar (or other special pedestal). It isn't
_the_ best strategy, but only one of a number of strategies which are each
best in their own way.

It is asymptotically the best strategy for a positive game when the bankroll
becomes sufficiently large.

If you were to take a million max-ER players and start them each with
$1000, and take a million min-risk players and start them each with
$1000, and let all the players play for a very long time, you would
find that the average _surviving_ max-ER player would have more
money than the average surviving min-risk player,

In fact, the average of *all* the max-ER players, including zeros for those
busted flat, will exceed the average of *all* min-risk players, though the
latter group will indeed include fewer bustouts.

Your max-ER bias is showing, and calling this "magic bullet strategies" seems
like an emotional appeal. The fact is, the max-ER promise of "more money"
has a clause hidden in the fine print -- you do get more money, but only if
your bankroll survives. The max-ER strategy does _not_ give you the best
chance for survival.

Not playing gives the best chance of survival. What Steve is calling
min-risk is a strategy optimal for minimizing the chance of bustout,
conditioned on a required bet size.

Q: Which is more important to you personally, having a higher
"wage" rate in $$$/hour, or having a higher probability of being
allowed to play as long as you wish, having your bankroll grow
without bound?

Your bankroll grows without bound playing any positive-ER strategy. The
question is, what is to be the tradeoff between risk of ruin and rate of
growth?

Those who choose $$$/hour should use max-ER strategy. Those
who chose better probability of getting rich should use min-risk
strategy.

No; for any fixed, finite monetary level defined as "getting rich", a
Kelly-optimal strategy minimizes the expected time to reach that goal.

Those bemused by this wrangling can be reassured that in most cases, for
most players, the differences between strategies is minor to nonexistent,
and the real effects on the chance of going busted, or the expected time to
reach a goal, are even less noticeable.

···

--
Randy Hudson

Randy Hudson wrote:

In fact, the average of *all* the max-ER players, including zeros
for those busted flat, will exceed the average of *all* min-risk
players, though the latter group will indeed include fewer bustouts.

Randy,

You've touched on a point here that my mind's been idly toying with
since the start of this thread.

Your statement is the equivalent of saying that no matter the point in
time, total cash held by max-ER players will be expected to exceed
that of the min-risk players.

It's occurred to me that some plays might ultimately result in the
greater surviving pool of min-risk players generating a revenue stream
that overcomes that of the fewer max-ER players despite their superior
ER - ultimately surpassing them in total expected cash on hand.

(I can see that if the players are able to escalate their play
denomination as bankroll warrants it, this possibility falls by the
wayside.)

At the risk of starting another tedious thread, can you give me some
insight here?

- Harry

Steve Jacobs wrote:
> I don't believe it is generally correct to change playing strategy based
> on the size of your bankroll. One playes max-ER because the objective
> calls for it, not because the bankroll size "calls for" or "allows"
> max-ER strategy.

"Correct" in pursuit of what goal? If you are playing a positive game, and
your goal is to maximize how much you expect to gain over any fixed number
of plays, then a max-ER strategy will achieve that (unless the bankroll is
very low relative to the advantage and variance). If your goal is to
minimize the chance you will eventually go broke, the best strategy is to
not play.

The "usual" min-risk goal for positive games is to minimize RoR which
really means to maximize the probability that you will play indefinitely
while your bankroll grows without bound. In both this case and the
max-ER case, the optimal strategy is fixed and doesn't change due
to bankroll.

Mixing a min-risk strategy with small bankrolls and a max-ER strategy
with large bankrolls gives a results that yields less ER that using
max-ER alone, and yields greater risk than using min-risk alone.

There are a number of goals that are equivalent to one another, however,
that call for slightly different strategies depending on the bankroll.
Those are the Kelly criterion goals, and they include minimizing the
expected time to reach a particular level of winnings, or maximizing the
chance that you end up with more winings than any other prestated strategy
over a sufficiently large number of trials.

This type of strategy change is somewhat artificial, and imposed by
the constraint that bet size has to match the units accepted by the
machine. Kelly betting seeks to wager the same fixed fraction of
bankroll on every play, and if the machines allowed arbitrary bet
sizes so that this would be possible, then the resulting playing
strategy would once again become unchanged as the bankroll
varies, and that strategy would differ from both min-risk and
max-ER.

Other types of boundary conditions can also call for changes in
strategy. That is why I qualified my statement by saying "generally
correct" rather than "always correct".

For these strategies, as the bankroll falls, the strategy becomes
increasingly risk-averse, as long as it retains positive expected
log-bankroll change. If the bankroll falls enough, that optimal strategy
will be not to play.

Agreed -- this is another type of boundary condition, based on
a minimum allowed wager.

If the bankroll falls enough, that optimal strategy
will be not to play.

The Kelly optimal strategy is "don't play". Your probability of
having the bankroll grow unbounded is non-zero if one continues
to play. Based on your discussion here, you seem to have bought
into the idea that "Kelly is everyting". I don't subscribe to that
theory (and yes, I have Thorp's paper).

If the bankroll grows enough, that optimal strategy will be the
max-ER strategy.

One would try to change unit sizes before that happens, but once
you've reached the largest unit size, and the bankroll grew to a
point where the unit size is a tiny fraction of the proper Kelly wager,
the strategy would change as you describe, for a player who
wishes to maximize log(bankroll). But at that point, you aren't really
doing Kelly betting any more, because you're constrained by unit
size.

In between, the strategy you refer to as
min-risk will be optimal for some range of bankroll, but except in a few
degenerate cases, it won't be the "best" strategy independent of bankroll.

From a Kelly perspective, what you say is true. From the perspective
of a player who seeks only to minimize RoR, rather than minimizing expected
log(bankroll), the min-risk strategy remains optimal for all bankroll sizes.

> One plays min-risk because they want the highest probability of having
> their bankroll last until a goal is reached. Min-risk does this best
> whether the bankroll is a single unit or a million units. This remains
> true whether the goal is "play indefinitely" as per risk of ruin, or
> the goal is "multiply initial bankroll by N' or just "build bankroll up
> to G units."

Well, if the time allowed is finite, these latter goals are
Kelly-equivalents; a bankroll-sensitive strategy will dominate the
bankroll-insensitive min-risk strategy.

I disagree. Kelly results in general are asymptotic, based on assumption
that time is unlimited. Kelly results dominate "eventually". In my opinion,
when time is finite, many Kelly results crumble like a house of cards, and
fixed fraction wagering morphs from a beneficial guiding principle to a
set of shackles that prevent one from devising an strategy that is truly
optimal for the finite boundries of realistic problems.

> 3) If the min-risk strategy says that the required bankroll is N units,
> no other strategy can give a smaller required bankroll for the same goal.

No other fixed strategy can, but a strategy that adapts to the current
bankroll level may.

Agreed, especially in games like VP. As the bankroll approaches the
goal, payoffs that would overshoot the goal can be treated as if they
are smaller than they really are, and the strategy adjusted accordingly.

> max-ER doesn't belong on an altar (or other special pedestal). It isn't
> _the_ best strategy, but only one of a number of strategies which are
> each best in their own way.

It is asymptotically the best strategy for a positive game when the
bankroll becomes sufficiently large.

I'll now turn your words against you -- "best" in pursuit of what goal?

I think you may be using Kelly as a "golden standard" for judging when
other strategies are "appropriate". I personally believe that is flawed
concept, and that too much is being read into the interpretation of
Thorp's Kelly paper. Kelly is a terrific model if you live forever and
are never confined by minimum/maximize bet sizes. It breaks down
for goals that are inherently finite in nature, such as maximizing
the probability of growing a bankroll to a fixed target size.

> If you were to take a million max-ER players and start them each with
> $1000, and take a million min-risk players and start them each with
> $1000, and let all the players play for a very long time, you would
> find that the average _surviving_ max-ER player would have more
> money than the average surviving min-risk player,

In fact, the average of *all* the max-ER players, including zeros for those
busted flat, will exceed the average of *all* min-risk players, though the
latter group will indeed include fewer bustouts.

I originally started to claim just the opposite, then backed off because I
decided I coun't easily prove it if challenged. If you can show this using
_arithemetic mean_ for the average, I'd appreciate seeing the derivation.
If you're making this claim based on Kelly results, then I would respond
by saying "that only holds if you force the use of geometric mean for
the averaging".

A large number of players who overbet their bankrolls by less than 2X
will as a group win more money than an equal number of players who
always bet the perfect Kelly fraction. The winnings will tend to be
more concentrated into the hands of a smaller number of players, but
the group _will_ have more money.

> Your max-ER bias is showing, and calling this "magic bullet strategies"
> seems like an emotional appeal. The fact is, the max-ER promise of "more
> money" has a clause hidden in the fine print -- you do get more money,
> but only if your bankroll survives. The max-ER strategy does _not_ give
> you the best chance for survival.

Not playing gives the best chance of survival. What Steve is calling
min-risk is a strategy optimal for minimizing the chance of bustout,
conditioned on a required bet size.

Correct. That's why I call it min-risk instead of zero-risk.

The zero-risk strategy is really boring.

> Q: Which is more important to you personally, having a higher
> "wage" rate in $$$/hour, or having a higher probability of being
> allowed to play as long as you wish, having your bankroll grow
> without bound?

Your bankroll grows without bound playing any positive-ER strategy.

True, but the _probability_ of having that happen always has bounds.

The
question is, what is to be the tradeoff between risk of ruin and rate of
growth?

> Those who choose $$$/hour should use max-ER strategy. Those
> who chose better probability of getting rich should use min-risk
> strategy.

No; for any fixed, finite monetary level defined as "getting rich", a
Kelly-optimal strategy minimizes the expected time to reach that goal.

Randy, this discussion isn't about Kelly, and never has been, largely
because VP isn't a very good fit for the Kelly model. At best, Kelly
can only be applied to the game in a very crude way, and a large
fraction of VP players are using a small enough bankroll that the
minimum unit size is a major impediment to subdividing the bankroll.

Those bemused by this wrangling can be reassured that in most cases, for
most players, the differences between strategies is minor to nonexistent,
and the real effects on the chance of going busted, or the expected time to
reach a goal, are even less noticeable.

Translation: For professionals with an adequate bankroll, none of this
matters much.

But declaring the difference "minor to nonesistent" doesn't make it so,
especially for those who aren't playing from your professional perspective.

···

On Monday 23 February 2004 06:34 am, ime@panix.com wrote:

Steve Jacobs wrote:

Steve Jacobs wrote:

> If you were to take a million max-ER players and start them each with
> $1000, and take a million min-risk players and start them each with
> $1000, and let all the players play for a very long time, you would
> find that the average _surviving_ max-ER player would have more
> money than the average surviving min-risk player,

In fact, the average of *all* the max-ER players, including zeros for those
busted flat, will exceed the average of *all* min-risk players, though the
latter group will indeed include fewer bustouts.

I originally started to claim just the opposite, then backed off because I
decided I coun't easily prove it if challenged. If you can show this using
_arithemetic mean_ for the average, I'd appreciate seeing the derivation.
If you're making this claim based on Kelly results, then I would respond
by saying "that only holds if you force the use of geometric mean for
the averaging".

A large number of players who overbet their bankrolls by less than 2X
will as a group win more money than an equal number of players who
always bet the perfect Kelly fraction. The winnings will tend to be
more concentrated into the hands of a smaller number of players, but
the group _will_ have more money.

I'm also a Kelly believer and this is intruding upon what I've always
regarded as sacred. One of the first things I learned about the Kelly
Criterion, long before I learned how to derive it, is that overbetting
one's bankroll by 2 times will yield an average result of breaking
even. Without knowing how to calculate it, but just plugging numbers
into a certain simple example by trial and error, I estimate that
overbetting one's bankroll only up to roughly 1.618 times will win
more money than optimal betting will. Past that, less money will be
won. Overbetting by 1/3 appears to maximize money won (which I had
never realized before and which begs the question of what the Kelly
Criterion maximizes, which I've never understood). Am I
misunderstanding your example? I assume you meant that the first
group of players may overbet their bankroll as close to 2X as they
want. Did you mean that?

···

On Monday 23 February 2004 06:34 am, ime@panix.com wrote:

Steve Jacobs wrote:

>A large number of players who overbet their bankrolls by less than 2X
>will as a group win more money than an equal number of players who
>always bet the perfect Kelly fraction. The winnings will tend to be
>more concentrated into the hands of a smaller number of players, but
>the group _will_ have more money.

I'm also a Kelly believer and this is intruding upon what I've always
regarded as sacred. One of the first things I learned about the Kelly
Criterion, long before I learned how to derive it, is that overbetting
one's bankroll by 2 times will yield an average result of breaking
even.

That is "true" but you have to be very precise in your definitions
of "average" and "breaking even".

Kelly betting seeks to maximize average log(bankroll), and this
is mathematically equivalent to maximizing the geometric mean
of the outcome.

Overbetting by 2X causes the _geometric_ mean to equal one,
when all possible outcomes are averaged _geometrically_.
However, if the apply an arithmetic mean over the exact same
outcomes, you find that the bankroll has grown exactly as predicted
by the game ER.

Without knowing how to calculate it, but just plugging numbers
into a certain simple example by trial and error, I estimate that
overbetting one's bankroll only up to roughly 1.618 times will win
more money than optimal betting will. Past that, less money will be
won. Overbetting by 1/3 appears to maximize money won (which I had
never realized before and which begs the question of what the Kelly
Criterion maximizes, which I've never understood).

I'm baffled as you what kind of averaging you would be doing to
reach this conclusion.

Am I
misunderstanding your example? I assume you meant that the first
group of players may overbet their bankroll as close to 2X as they
want. Did you mean that?

I believe there is a lot of confusion as to what Kelly betting really does,
even amount those who have spent a lot of time studying Kelly.

Consider several different groups who all use fixed-fraction betting, but
each group using a different fraction. For positive EV games, betting
any fraction larger than zero will cause the number of dollars in each
group to grow over time, and since each group wagers a fixed fraction
of each player's bankroll at each betting opportunity, the bankroll of
each group will grow exponentially. Any "very large" group that bets
more than 1X Kelly will see the _group_ bankroll grow at an exponential
rate that is higher than the growth rate of the _group_ bankroll of the
Kelly group. This remains true even when you go beyond 2X Kelly,
but this is in terms of total dollars.

So, Kelly betting does _not_ win "the most money" if you look at
all the dollars won by each _group_. What Kelly does is control
the distribution of dollars within each group in such a way that the
geometric mean of the individual bankrolls will be as large as
possible.

One result of this is that the median player of the Kelly group (the guy who
beats 50% of the other Kelly players and is beat by the other 50% of Kelly
players players), will fare better than the median player in any of the other
groups. That is one of the results stated in Thorp's paper.

Another consequence is that if you let all the groups keep playing for
a long time, the Kelly group will eventually "dominate" any other group.
One big misconception is that "dominate" means "wins more money,
as a group". What dominate really means is that if you line up the players
in the Kelly group, in order of bankroll, and set them side by side with
the players of another group, then the percentage of Kelly players who
are ahead of the corresponding player in the other group will become
larger an larger as the group play longer and longer. So, if you compare
the Kelly group with _any_ group that uses a larger betting fraction, and
pick a Kelly player at random, the corresponding player in the other
group will be likely to have less money. What this means is that even
though the overbetting group wins more total dollars, those dollars
tend to become concentrated in the hands of a smaller percentage of
players.

Now, if a group bets less than 2X Kelly, the geometric mean of their
bankrolls will grow. If the group bets more than 2X Kelly, the geometric
mean of their bankrolls will shrink even though the group as a whole
wins more money than the Kelly group. As long as betting is less than
2X Kelly, the median player can expect to come out ahead eventually,
and in fact the percentage of players who come out ahead will grow
as play continues. If the group bets more than 2X Kelly, all the
enormous winnings of the group will end up in the hands of a small
(and shrinking) percentage of lucky player while the rest see their
bankrolls dwindle. But, as a group, the lucky few win much more
money than the entire Kelly group combined.

The confusion about Kelly comes about because the outcome is
usually described in a way that only looks at the cases where the
Kelly player comes out ahead. It is pointed out that the percentage
of "Kelly winners" increase, approaching unity as more hands are
played. While this is true, it ignores a vast fortune in winnings that
are concentrated into the hands of those players who beats their
Kelly counterparts.

In short, betting Kelly doesn't "win more money" than betting 1.1X
Kelly, but it does help distribute the winnings in a way that many
feel is "more equitable". However, the "superior" performance
is always something that is "eventual" if the players continue
until the Kelly group ultimately "looks better". This implies that
Kelly is well suited for looking at the long-run, but not necessary
well suited for situations that are inherently bounded in bankroll
size and/or in time.
than

···

On Tuesday 24 February 2004 06:25 am, Tom Robertson wrote:

Steve Jacobs wrote:

Steve Jacobs wrote:

>A large number of players who overbet their bankrolls by less than 2X
>will as a group win more money than an equal number of players who
>always bet the perfect Kelly fraction. The winnings will tend to be
>more concentrated into the hands of a smaller number of players, but
>the group _will_ have more money.

I'm also a Kelly believer and this is intruding upon what I've always
regarded as sacred. One of the first things I learned about the Kelly
Criterion, long before I learned how to derive it, is that overbetting
one's bankroll by 2 times will yield an average result of breaking
even.

That is "true" but you have to be very precise in your definitions
of "average" and "breaking even".

Kelly betting seeks to maximize average log(bankroll), and this
is mathematically equivalent to maximizing the geometric mean
of the outcome.

Overbetting by 2X causes the _geometric_ mean to equal one,
when all possible outcomes are averaged _geometrically_.
However, if the apply an arithmetic mean over the exact same
outcomes, you find that the bankroll has grown exactly as predicted
by the game ER.

Without knowing how to calculate it, but just plugging numbers
into a certain simple example by trial and error, I estimate that
overbetting one's bankroll only up to roughly 1.618 times will win
more money than optimal betting will. Past that, less money will be
won. Overbetting by 1/3 appears to maximize money won (which I had
never realized before and which begs the question of what the Kelly
Criterion maximizes, which I've never understood).

I'm baffled as you what kind of averaging you would be doing to
reach this conclusion.

The example I was applying this to was a proposition in which 99/x of
one's bankroll would be lost 50% of the time and 101/x of one's
bankroll would be won 50% of the time. Setting x to 9999 maximizes
the formula ((1+(101/BR))^.5)*((1-(99/BR))^.5). Let this formula be
called "F." Following your lead of seeking to maximize dollars won
leads to maximizing the formula (F-1)/BR by setting x to 7499.
Setting x to 4999.5 brings F to 0 and setting x to 6180 brings
(F-1)/BR to roughly the same as setting x to 9999 does.

<snip very enlightening discussion about Kelly>

···

On Tuesday 24 February 2004 06:25 am, Tom Robertson wrote:

Specifically, I'd like the min-risk strategy for 9/6 JoB.

Dave

Tom -

Steve J is probably formulating a brilliant reply to your post, in
which case you should probably ignore my likely naive
comments/questions. You REALLY lost me.

How do X and BR relate? Is X = (BR/BET)*100?

Seems like BR=9999 not X=9999 maximizes
((1+(101/BR))^.5)*((1-(99/BR))^.5)

Seems like setting BR (x?) to 4999.5 brings F to 1, not 0. Did you
mean that it brings (F-1)/BR to 0?

I don't understand what (F-1)/BR is supposed to represent. You seem
to define it as somehow representing "dollars won". Could you
elaborate on that?

Seems to me that by a simple definition "dollars won" is maximized by
maximizing
(101/X)*0.5 - (99/X)*0.5 = 1/X
We maximize this by setting X as low as possible, i.e. bet it all.

AJ

Tom R wrote:
<snip>

The example I was applying this to was a proposition in which 99/x

of

···

one's bankroll would be lost 50% of the time and 101/x of one's
bankroll would be won 50% of the time. Setting x to 9999 maximizes
the formula ((1+(101/BR))^.5)*((1-(99/BR))^.5). Let this formula be
called "F." Following your lead of seeking to maximize dollars won
leads to maximizing the formula (F-1)/BR by setting x to 7499.
Setting x to 4999.5 brings F to 0 and setting x to 6180 brings
(F-1)/BR to roughly the same as setting x to 9999 does.

Steve Jacobs wrote:

<snip>

So, Kelly betting does _not_ win "the most money" if you look at
all the dollars won by each _group_. What Kelly does is control
the distribution of dollars within each group in such a way that the
geometric mean of the individual bankrolls will be as large as
possible.

One result of this is that the median player of the Kelly group (the guy who
beats 50% of the other Kelly players and is beat by the other 50% of Kelly
players players), will fare better than the median player in any of the other
groups. That is one of the results stated in Thorp's paper.

It sounds like what the Kelly criterion maximizes could be summed up
as winnings assuming that there's no fluctuation from expected in the
frequency of any possible result. This doesn't strike me as
necessarily optimal, either. It radically differs from expected
value, even long-term, since it undervalues good luck so much. It's
as if it knew that money has diminishing marginal utility (the only
theoretical reason I've ever been aware of not to overbet), but I'm
skeptical that accurately accounting for that would correlate very
well to the Kelly Criterion.

<snip>