def rfstockpicker_backtest(database,prediction_date,rolling_window,
                           nb_months,n_estimators,n_jobs):
    database = database
    results={}
    for i in range(0,nb_months):
        d = datetime.datetime.strptime(prediction_date, "%Y-%m-%d")

        d2 = d - dateutil.relativedelta.relativedelta(months=i)
        d3 = d - dateutil.relativedelta.relativedelta(months=i+1)
        d4 = d3 - dateutil.relativedelta.relativedelta(years=rolling_window)    ##Creating a rolling window interval

        test_set = database[database['decision_date'].isin(pd.date_range(d2, d2))]
        training_set=database[database['decision_date'].isin(pd.date_range(d4, d3))]

        # Labels are the values we want to predict
        train_labels = np.array(training_set['win'])
        # Remove the labels from the features
        columns_to_drop=['win','sprtrn','gvkey','datadate','conm','trt1m','decision_date','tic']
        train_features= training_set.drop(columns_to_drop, axis = 1)
        # Saving feature names for later use
        train_feature_list = list(train_features.columns)
        # Convert to numpy array
        train_features = np.array(train_features)

        test_labels = np.array(test_set['win'])
        # Remove the labels from the test features
        test_features= test_set.drop(columns_to_drop, axis = 1)
        # Convert to numpy array
        test_features = np.array(test_features)

        # Import the model we are using
        from sklearn.ensemble import RandomForestRegressor
        # Instantiate model
        rf = RandomForestRegressor(n_estimators = n_estimators, random_state = 42,n_jobs=n_jobs)
        # Train the model on training data
        rf.fit(train_features, train_labels);

        # Use the forest's predict method on the test data
        stock_predictions = rf.predict(test_features)
        result_set=test_set
        result_set['predictions']=stock_predictions
        results[i]=result_set

       # Get numerical feature importances
        importances = list(rf.feature_importances_)
        # List of tuples with variable and importance
        feature_importances = [(feature, round(importance, 2)) for feature, importance in zip(train_feature_list, importances)]
        # Sort the feature importances by most important first
        feature_importances = sorted(feature_importances, key = lambda x: x[1], reverse = True)

    return results,feature_importances,importances,train_feature_list