Do you have internal function(market prediction, internal prediction, mana available, current potential % income, alternative markets with higher % income, influence on your calibration, liquidity)? What are the weights?
5-20 mana for honest answer
Related questions
25 mana in markets I’m moderately confident in, 50 where I’m very confident. More than that where I think the market is wildly wrong, or if it’s almost certain to resolve very quickly and I can turn a quick few mana. Less than that when someone is offering a boost I want to claim (I’ll usually just invest the boost) or when I want to bookmark a position for later review.
Means I’m rarely heavily exposed on a single market and my portfolio growth may be slower but should be more stable than going heavy on riskier investments.
25 and 50 use to be 15 and 30, and will probably increase again when I find myself consistently with Mana to spare.
Whatever Ṁ i need to bet to make the market within 5% of my prediction, with the upper limit of 5k (1/6 of portfolio). Except for shenanigans like "AI doom" markets. If the market is way off my prediction, i don't bet it down/up all at once, but split my bets over multiple days with limit orders. If i run out of mana, i close some less confident positions that i can close without taking a loss.
Not that I use it consistently, but are you familiar with the Kelly bet size?
I want to know everyone's answer to this so I will add on.
I have tried, randomly for different questions, different "strategies" (not strategic at all):
'The lazy way': some fixed, categorical amounts that is decided on too many random factors, but also (internally monologued) rational factors like confidence and interest in a question. These are usually smaller bet sizes ~1-25 mana (up to 100 Mana). These type of bets are made rashly usually and are for low bet sizes.
'The speculative and risky way': the market % is way off and I am (arrogantly) confident that I will be right and the market is wrong. Per question I will consider bet sizes in this case as a % of my Balance (10%-50%). These type of bets usually involve questions that are not quantitatively inclined but that rely on my intuition for and I find very interesting (less frequent, higher amount).
'Quantitative bet sizing': I've used a variation of the Kelly criterion a couple of times when I have a calculation for the question % and it is sufficiently different then the market's %. What I would do is also imagine how many more times I will update before I think the question will resolve and divide the bankroll for the criterion by the number of updates I expect to do. If I miss updates then I will adjust later on in future bets. These are usually 100 M bank roll bets over some period. Sometimes I am betting quite a bit more depending on my confidence in my analysis. These types of bets involve some type of required learning and upskilling, as well as a question that fuels my interest (more frequent, medium size bets); this strategy is generally low-reward with many updates/bets. I have considered (but not done so) for speculative purposes the risk-reward might in general be better matched to exponentially reducing my bets over time for this strategy, especially when it is the case that I will update regularly; this is too much calculation for the purpose of only chasing Mana (maybe if I wrote a program to do this with the API in the future...).
'Equal bet update corrections': Given a fixed bankroll (10-100 mana) and a clear date when the question will resolve, I will just divide the bankroll by the number of days and bet on each update whatever way I think moves the market in the correct direction. This is fast and simple to compute, but is generally low-reward -- for updates near resolution time is high risk and low reward. I've used these types of bets for questions that % usually monotonically increase/decrease in one direction over time; since they do not clamp early, they often have a feature where it is usually the case where there is also a probability of an anomaly that resolves the question early (death markets would be an example).
'Unequal bet update corrections': In some questions where I have quantitative data and have created an analysis program to help update over a period of time, I've modified my bets such that I first estimate how much mana it will currently take to move the market question % to my estimated %, and then use that amount of mana as the bankroll (this can only be done in certain circumstances, but I've done it even if the bankroll exceeds my balance). For example: including my confidence in updating correctly and regularly, such as if I think there is a 90% of updating correctly and regularly for a market that resolves in 30 days and it currently takes 1000 mana to move the market to my estimate, for the first day I would do (0.9^30) * 1000, and then decrement 30 for each successive day that I update as well as change the mana needed to move the market (this is time consuming). I feel this synthetic approach takes into account my own probability estimate of the question, the market probability, and my own estimate of how well I will do in each update. I would like to know if there is another approach that does something similar to this, but is more rational as this calculation probably too heavily discounts early updates.
'Pure interest/fun bets':
Questions that I want to see happen or don't want to see happen. Token bets < 10 Mana (for fun) to express interest in the question.
I bet 1 mana on every market because I am a slave to my own market
https://manifold.markets/f/i-will-bet-on-all-your-makets-if-yo
Approach 1.
I read the description and form my own prediction Metaculus style x%. Then I look at y% the current market value. Bet the difference.
If I predict 90% and the market is at 60%, then I bet 30 mana on YES. If I predict 2% and the market is at 3%, I bet 1 NO.
It has no maths behind it, it just converts higher dissonance into more mana.
Approach 2.
If some market theme in my opinion gets "reverse half-life updates" (news appear more and more frequently, like a mirrored half-life function), then I sell other positions and fully invest (50% of all available to me mana) into the market.
I might try in the future:
mana_to_bet = (predicted% - market%) * current_balance.
If I predict 90% and the market is at 60%, then I bet 30% of my balance