# Increase the probability of correct prediction using multiple regression

First off let me begin by saying that I'm brand new to statistics and I would appreciate it if you could dumb down any answers for my problem.

I am trying to create a general prediction of how much a stock will go up in value in the following 7 days. What I have done is get a list of all percentages that the stock went up for each 7 day period, I got a list of the inputs and performed a multiple regression formula and received the coefficients for each input (3 inputs). I tested the coefficients for each stock and they are between 0.9 and 1 which I understand is very good.

Predicted Return = ((Rating1 * Rating1Coef) + (Rating2 * Rating2Coef) + (Rating3 * Rating3Coef)) * rsquared

I then went back over the list of all percentages that the stock made for each 7 day period and compared it to my predicted value using the inputs and the coefficients and below is a list of stocks and the percentage of how often my prediction was correct (i.e. my prediction was lower than how much the stock actually made)

Symbol | Probability of Predicted < Actual | Avg Pct Difference of Predicted vs Actual
JNUG  | 72.26% | -44.80%
TLL  | 6.22% | -58.44%
TKMR  | 14.45% | -44.64%
ENRJ  | 14.64% | -48.18%
GENE  | 21.90% | -17.39%
ZIOP  | 14.06% | -25.15%
DWTI  | 52.46% | 22.78%
DGAZ  | 93.41% | -107.66%
SQQQ  | 5.71% | -63.30%
ASPS  | 17.18% | -81.11%

1. Is there any information that you need to know that I'm leaving out?
2. What general steps can I take to increase the probability of choosing an amount that is close to the actual amount?
3. Since my rsquared values are all fairly high, this just means that the predicted value will come close to the average amount that the stock made with similar inputs right?